This research focuses on multiple-input, multiple-output (MIMO) signal separation in environments with changing channels, such as mobile communications. When signals are separated by combining weights that are fixed in a packet or until the next reference symbol, the orthogonality between interference signals and combining weights becomes broken due to the movement of the receiver, resulting in interference leakage. To mitigate such degradation, a null-space expansion technique is employed, assuming that there are extra antennas at the receiver side. This technique improves channel time-varying immunity by using extra receiving antennas to form a null-space that suppresses interference among streams in MIMO or inter-user interference (IUI) in multi-user MIMO. In this paper, the performance advantage of the null-space expansion technique is revealed through in-lab and field measurement campaigns.

In data centers (DCs), both inter- and intra-DC flows share high-speed links in Data Center Networks (DCNs). These two types of flows have significantly different Round-Trip Times (RTTs). When congestion occurs in a DCN, the congestion signal is first detected by the intra-DC flow, leading to unfairness because of fast reaction to the congestion. This technical issue is particularly critical for the receiver-side DCN, where the congestion point is distant from the sender. To address this issue, we have proposed a new approach of proxy gateway named DC-PEP with ECN, which divides the end-to-end congestion control loop at the gateway node in receiver-side DCN. DC-PEP can balance the congestion loop inside the receiver-side DCN. Furthermore, ECN-Echo notifications at DC-PEP avoids excessive queueing in the proxy. However, just a simple ECN mechanism might cause overshoot of queue size at DC-PEP due to delays in ECN-Echo notifications caused by WAN propagation delays. To resolve this technical issue, we propose DC-PEP with Gradient-based ECN (G-ECN), utilizing a new ECN congestion control mechanism. Early ECN-Echo notifications facilitated by gradient of queue size can contribute to stabilization of queue size. Our simulation results demonstrate that the proposed DC-PEP with G-ECN enhances queue stability and improves fairness between inter- and intra-DC flows.

Future carrier networks for 6G are expected to verify the performance across heterogeneous networks integrating multiple technologies. This integration requires effective verification of network performance under unpredictable conditions. In response, a node modeling method has been proposed to estimate actual network node performance before deployment. However, developing such models based on machine learning algorithms is challenging due to the high cost and difficulty of collecting extensive real-world data under varying conditions. Alternatively, fine-tuning the model using data from data-rich domains can enhance estimation accuracy; however, it also raises operational costs by necessitating careful selection of relevant data and parameter tuning. Therefore, this study aims to improve the accuracy of router performance estimation using limited real-world training data while avoiding complex fine-tuning. We propose an attentive Retrieval-Augmented Generation (RAG) method with Large Language Models (LLMs) to build a model for estimating actual router performance by augmenting real-world data with the most relevant simulation data. The effectiveness of the proposed method is evaluated using software routers executed on commodity servers in a single-router scenario. We investigate how natural language processing techniques capture subtle feature relationships to enhance the accuracy of router performance estimation. Experimental results show that the proposed attentive RAG with LLM improves the estimation accuracy of actual router performance metrics such as throughput, packet loss rate, and packet delay compared to a baseline regressor without LLM and to one with LLM in a scenario with limited real-world training data.

Sparse code multiple access (SCMA) is a non-orthogonal multiple access scheme that can achieve high sum-rates. However, SCMA requires a large computational complexity for detection. In this paper, we propose a scheme that reduces the computational complexity required for detection in downlink SCMA systems. In our scheme, users’ constellation points are aligned on the I and Q axes and additional mutually orthogonal axes. Each of the four axes has an orthogonal axis, and the set of the constellation points on one axis is independent of the set on the orthogonal axis. Therefore, by considering each axis as an independent resource, we can treat one subcarrier resource as four distinct resources. This can be achieved by using a factor graph that we named the joint IQ factor graph. In this case, the number of multiplexed constellation points on the resource can be reduced. Furthermore, we introduce two types of SCMA codebooks that enable further complexity reduction using the joint IQ factor graph. One combines our scheme with a low-projection codebook, and the other reduces the size of codebook. We demonstrate that our new scheme can realize an SCMA detector with a modern low power microcontroller.

Using unmanned aerial vehicles (UAVs) for construction site management has been advancing to improve operational efficiency. However, construction sites often have poor radio environments due to obstructions from structures and machinery, and these environments change dynamically as construction progresses. Clarifying the radio propagation characteristics under these changing conditions is essential to ensuring the safe and reliable operation of UAVs. This paper investigates the received signal strength characteristics of UAVs and controllers through field measurements in environments without structures and with significant structural presence. Without structures, we confirmed that the received power attenuates proportionally to the distance raised to the 2.1 power, consistent with free-space propagation. Measurements were conducted at multiple stages of construction progress in environments with structures, revealing that when a line-of-sight (LOS) path cannot be maintained, the propagation loss increases in proportion to the distance raised to powers between 2.1 and 2.5. The path loss exponent increased as the construction advanced, indicating that structural growth significantly affects UAV communication quality. These findings suggest that the RMa-NLOS propagation model, which is defined for rural macro environments, is applicable for estimating propagation loss in similar construction site scenarios.

The emergent direct-to-cell low Earth orbit (LEO) satellites are designed to address the coverage holes of terrestrial networks, providing connectivity for unconnected global users. Such LEO satellite constellations are considered an essential part of the fifth-generation (5G) and beyond non-terrestrial network. However, the movement of LEO satellites triggers a high number of mobility events, and an increase in these events could lead to signaling storms and potential service failures. To enhance mobility performance, the Third Generation Partnership Project (3GPP) suggested that additional triggering criteria based on the satellite’s trajectory and the UE’s location should be considered. After that, to mitigate signaling storms caused by the high mobility of LEO satellites, a location-based group handover (LGHO) protocol was proposed to allow LEO satellites, acting as full gNB (next Generation NodeB in 5G) nodes, to flexibly configure user equipments (UEs) to perform group/individual handover based on real-time situations. However, in cases of high UE density, the performance of LGHO could be worse than that of location-based baseline handover (LBHO). In this paper, we propose a nomadic Integrated Access and Backhaul (IAB)-based solution to mitigate signaling storms in 5G and beyond LEO satellite networks. In the proposed solution, nomadic IAB-nodes are dynamically and selectively deployed in areas with high UE density, and then LEO satellites function as intermediate IAB-nodes, connecting nomadic IAB-nodes and an IAB-donor. The evaluation results show that the proposed solution outperforms LGHO and LBHO in terms of the handover success rate and the number of signaling messages received by each satellite.

This invited paper presents our vision of Artificial Intelligence (AI)-Native radio transceiver design for next-generation (6G) mobile communication systems by integrating AI models into the conventional radio signal processing pipeline. The proposed design and architecture can efficiently handle wireless signal processing, enabling reliable and high-performance mobile communications. Unlike traditional transceiver designs, which are agnostic to the environment, the proposed AI-Native approach leverages dynamic learning of existing regularities and patterns in the wireless environment (e.g., channel, hardware nonlinearity, traffic, and user mobility models) to improve system efficiency. We demonstrate how AI can enhance the performance of key functions such as channel estimation, beamforming, link adaptation, and phased-array antenna calibration. Unlike prior work, we emphasize the practical deployment of AI models in a millimeter-wave communication testbed and demonstrate a realistic training approach based on real-world data, moving beyond the constraints of synthetic datasets. Finally, we present AI-enabled extended reality as a promising 6G application and highlight its requirements and the challenges it poses on the AI-Native radio access network.

This paper proposes a comprehensive framework for constructing smart wireless aerial networks to support high-fidelity digital twin (DT) systems. To meet the stringent data rate demands of airborne Light Detection and Ranging (LiDAR) sensing, we analyze the required throughput for point cloud transmission and design a millimeter-wave (mmWave) aerial link budget model. A software-defined architecture integrating Network Function Virtualization/Software Defined Networking (NFV/SDN) is introduced to enable dynamic Unmanned Aerial Vehicle (UAV) orchestration, routing, and network slicing. To enhance robustness and efficiency, we propose two application-layer innovations: a multi-route redundant communication framework and a semantic image transmission protocol using deep joint source-channel coding (DJSCC) with feature-based elastic compression. Furthermore, we implement a multi-agent reinforcement learning strategy to enable autonomous UAV placement and relay network formation in dynamic environments. Simulation results demonstrate the proposed system’s scalability, adaptability, and potential to enable reliable and efficient DT construction across diverse deployment scenarios.

In this paper, we propose a novel SI cancellation method in the presence of a fractional delay path for in-band full-duplex (IBFD) communications, considering its application in Bluetooth low energy (BLE). IBFD facilitates simultaneous transmission and reception of signals within the same frequency band. This technology offers substantial advantages over half-duplex (HD) communications by doubling the spectral efficiency and eliminating the need for timing control during signal transmission and reception. However, SI cancellation is essential for the implementation of IBFD. Traditional SI cancellation methods, such as the active SI canceller (ASIC), generate and subtract a replica of the SI signal from the received signal. However, generating this replica requires accurate channel state information (CSI), which increases communication overhead. Therefore, the differential ASIC (DASIC), which does not require prior CSI, has been proposed. However, the presence of a fractional delay path results in residual SI, which makes cancellation insufficient. To address this inconvenience, we propose a novel method to blindly remove the residual SI components induced by the delayed path by adjusting the differential delay controlled inside the DASIC according to the occurrence pattern of its own transmission signal. In addition, an appropriate maximum a-posteriori probability (MAP) detector is designed for minimum shift keying (MSK) and Gaussian MSK (GMSK) signals based on DASIC outputs. Finally, computer simulations confirm the effectiveness of the proposed method for BLE in terms of bit error rate (BER) performance.

This paper proposes a novel Bluetooth low energy (BLE)-angle of arrival (AoA)-based indoor localization method using neural networks (NN) to improve accuracy. Existing BLE-AoA localization approaches are typically based on triangulation or least squares (LS) algorithms, but the accuracy degrades due to noise, multipath effects, non-line-of-sight (NLOS) conditions and other factors. Among these, the weighted LS (WLS) algorithm has been introduced to mitigate the impact of unreliable AoA measurements by adjusting weight coefficients. However, the propagation characteristics of radio waves in indoor environments are highly complex, making it difficult to analytically determine optimal weight coefficients. To address this, we introduce a data-driven tuning approach for WLS using NN models that take AoA data as input and output optimized weights. In this paper, two NN structures are evaluated: a multi-layer perceptron (MLP) and a long short-term memory (LSTM) model. Experimental results demonstrate the effectiveness of the proposed localization method in terms of localization accuracy in two different environments, assuming moving object tracking.

Traditional adaptive beamformers often exhibit performance degradation under model mismatches or limited sample scenarios. To address these limitations, this paper proposes a robust adaptive beamforming algorithm that integrates diagonal loading and interference-plus-noise covariance matrix (INCM) reconstruction. Firstly, a dynamic mechanism is developed to determine the diagonal loading factor based on the signal-to-noise ratio (SNR), eigenvalue dispersion, and the number of snapshots. Secondly, the INCM is reconstructed by integrating the optimized spatial power spectrum over the non-target signal region. Finally, a steering vector solution method based on a two-step optimization process is derived and formulated to optimize the steering vector of the desired signal (DS) within its angular neighborhood. Simulation results demonstrate that the proposed algorithm effectively handles challenging scenarios, such as limited snapshots, large angular deviations, and multiple steering vector mismatches. It also consistently outperforms the traditional Capon beamformer and representative INCM reconstruction-based algorithms.

Information-Centric Networking (ICN) has emerged as a promising future network architecture by enabling efficient content retrieval, flexible routing, and enhanced support for data-centric services. To realize these advantages in practice, introducing IP-compatible ICN overlays into existing infrastructures — forming hybrid ICN/IP networks — has become an essential step and an important component of multi-protocol coexistence in future network architectures. Recent studies have shown that such hybrid networks leverage ICN’s intrinsic advantages in content delivery scenarios. However, the mismatch between ICN’s name-based and IP’s location-based protocols poses a challenge: ICN’s caching and multipath mechanisms conflict with IP’s strict end-to-end semantics. This protocol mismatch complicates resource scheduling and renders existing traffic-control frameworks ineffective, causing data retransmissions and reduced user throughput, thereby degrading QoS in content distribution scenarios. Therefore, we propose a traffic control strategy, TCS-II, for ICN/IP hybrid networks in content delivery scenarios. TCS-II is completely compatible with the characteristics of the existing ICN/IP hybrid architecture. By leveraging historical traffic information at the edge of the ICN network, it constructs a Lyapunov drift-plus-penalty framework to enable rate adaptation across heterogeneous protocols and to design a buffer management algorithm for ICN nodes. Additionally, by introducing a buffer-driven redundancy factor field into the scalable ICN packet structure, it assists ICN transmission nodes in adaptively allocating paths for flows, thereby minimizing ICN transmission costs. Simulation results show that TCS-II effectively reduces retransmissions and transmission costs. Compared to existing methods, user throughput is improved by 21.92% and 47.61%, respectively, effectively shortening flow completion times.

Remote Direct Memory Access (RDMA) is a key technology in Data Center Networks (DCN) owing to its low latency, high bandwidth, and Central Processing Unit (CPU) bypass capabilities. However, RDMA over Converged Ethernet version 2 (RoCEv2) is suitable for standard RDMA transmission, making it difficult to utilize the abundant parallel resources in DCNs. While multipath transmission enhances bandwidth, it introduces out-of-order (OoO) packets. Recent studies have confirmed that even a small fraction of OoO packets can significantly degrade RDMA throughput due to the Go-Back-N (GBN) retransmission mechanism. OoO events frequently occur in practical DCNs when flowlets are split too aggressively or packets are sprayed across multiple paths, making OoO handling a critical problem. Moreover, existing bitmap-based solutions suffer from low memory efficiency and high processing delays. To address these problems, this study proposes TRNIC, an optimized bitmap design for fast and efficient packet reordering. TRNIC adopts a triple-table structure for bitmap sharing and memory efficiency, an array structure for efficient random access, and scheduling isolation for concurrent Work Queue Elements (WQEs) within a single Queue Pair (QP). The proposed TRNIC innovatively presents a high-performance OoO packet handling solution based on bitmap for the first time. Compared with Mellanox CX5 and Xilinx ERNIC, our proposed TRNIC significantly improves RDMA throughput under OoO conditions, achieving nearly 97 Gbps, and greatly reduces Flow Completion Time (FCT) compared to GBN.

This study presents novel calibration techniques for a satellite receiving digital beamforming antenna, using ground calibration sources. The proposed technique utilizes calibration sources with known locations and cross-correlation vectors, which are the averages of the products of the array received signal vector and the received signal of the reference antenna element. The proposed method requires accurate information on the calibration source directions, and thus antenna pointing errors resulting from the satellite attitude fluctuation deteriorate the calibration accuracy. To address this issue, the method is extended to a blind calibration technique, in which amplitude and phase errors of the antenna and the antenna pointing error are estimated concurrently. The extended method also uses the cross-correlation vectors and incorporates a constrained minimization problem of the L₁-norm of the amplitude errors. Simulation results demonstrate the effectiveness of the proposed techniques.

Long-term rain attenuation statistics of the Ka and Ku band satellite signals obtained from 1988 to 2006 at Osaka Electro-Communication University in Neyagawa, Osaka, are investigated for the yearly variation of their attenuation characteristics. The ratio of Ka-band to Ku-band attenuation indicates yearly fluctuations of more than ±10% around the mean value in their equi-probability relationship, being considerably larger than the frequency scaling model of the ITU-R predictions. This study reveals that the deviation from the ITU-R frequency scaling model is primarily caused by the difference of raindrop size distribution (DSD) between the stationary fronts that give rise to the rainfall in the East Asian rainy season (Tsuyu or Baiu) or Autumn rain (Akisame) unique to Japanese climate. Also, the occurrence rate of summertime shower and typhoon is found to affect DSD. In addition, the deviation rate of the attenuation ratio from the ITU-R predictions is found to be correlated with the yearly time percentages of stationary fronts and typhoon. For practical satellite communications links, yearly and long-term time percentages and the rain margins required for specific time percentages are then presented to the CS rain attenuation actually measured and estimated by the frequency scaling method.

This paper proposes adaptive distributed detection schemes for terminal-collaborative multiple-input multiple-output (MIMO) systems, in which multiple wireless terminals jointly receive MIMO signals transmitted from a base station. To reduce collaboration traffic while maintaining transmission performance, an on-demand multiple detection terminals scheme is introduced, where detection terminals evaluate the reliability of their decision results and selectively request more reliable decisions from peer terminals within the collaboration group. Building on this, an on-demand adaptive multiple detection terminals (OAMDT) scheme is proposed, wherein detection terminals dynamically assume the role of helper terminals to forward their received signal waveforms when additional diversity is needed due to persistent decision errors. Both schemes employ frequency-domain iterative equalization using MMSE filtering and LDPC decoding, and utilize a residual error coefficient as a reliability metric to guide decision fusion and adaptive role assignment. Simulation results under various fading environments and SNR conditions demonstrate that while the on-demand multiple detection terminals scheme effectively reduces traffic overhead, the OAMDT scheme significantly enhances packet error rate performance with only a modest increase in collaboration traffic. The results further reveal diminishing returns in error performance when increasing the number of detection terminals, suggesting that adaptive role switching is particularly beneficial under constrained conditions such as overloaded MIMO scenarios. These findings indicate that the proposed adaptive collaboration mechanisms can enhance the efficiency and robustness of terminal-collaborated MIMO reception.

In this paper, we introduce the learned cooperative projected gradient descent (LCoopPG) algorithm for fully distributed multiple-input multiple-output (MIMO) signal detection. The proposed algorithm is a fully distributed algorithm, implementing the projected gradient method over a network, and incorporating a trainable step size and trainable weight parameters, which are tuned by deep unfolding. With the trained parameters, the computations of the proposed algorithm are fully distributed, eliminating the need for a central processor. Numerical experiments affirmed the superiority of the LCoopPG algorithm over the conventional minimum mean squared error algorithm in terms of detection performance. The experimental result underscores the efficacy of deep unfolding, indicating that proper tuning of the trainable parameters can significantly improve symbol error rate performance. Another notable advantage of the proposed algorithm is its scalability. Scaling the number of processing units (PUs) reduces the general purpose graphics processing units computational demands on each PU.

To address the security challenges of underwater wireless sensor networks (UWSNs) in complex and hostile environments, this paper investigates the physical-layer secrecy performance of the transmitter-receiver link, with particular emphasis on transmission optimization under eavesdropping threats. We formulate a non-linear, non-convex optimization problem that jointly allocates sub-channels and transmission power to maximize the secrecy rate while considering the inherent constraints of underwater acoustic communication. To efficiently solve this problem, we develop a deep reinforcement learning (DRL)-based resource allocation algorithm that integrates a prioritized experience replay mechanism, thereby enhancing learning efficiency and improving optimization performance. Extensive simulations are conducted to assess the convergence and adaptability of the proposed method under varying node densities and maximum transmission power levels. The results demonstrate that the proposed algorithm consistently achieves higher system security, greater stability, and faster convergence compared to several benchmark schemes.

In-band full-duplex communication has the potential to theoretically double spectral efficiency and is a promising next-generation communication technology. However, to address the severe self-interference arising from simultaneous transmission and reception over the same frequency band, the application of self-interference cancellation (SIC) techniques is essential. In this study, a digital-domain SIC approach is evaluated, considering practical imperfections in radio frequency (RF) circuits, including nonlinear distortion from power amplifiers, in-phase/quadrature imbalance, and direct-current offsets. An in-band full-duplex communication terminal was constructed using a software-defined radio and RF circuits, adopting a shared local oscillator architecture to reduce phase noise effects. A parallel Hammerstein canceller based on a low-intermediate-frequency architecture and a parallel Hammerstein model was implemented to mitigate nonlinear self-interference components. SIC performance was evaluated using orthogonal frequency-division multiplexing (OFDM)-modulated signals. A digital-domain SIC ratio exceeding 50 dB was achieved. These results demonstrate the effectiveness of the parallel Hammerstein canceller in a practical in-band full-duplex communication system employing OFDM signals.

This paper proposes a method that selects a set of transmission reception points (TRPs), a TRP group, to be connected for each user equipment (UE) to maximize the system throughput defined based on the generalized mean of the user throughput in a downlink distributed multi-input multi-output (MIMO) system. As a realistic assumption, the TRP groups to be selected are pre-determined by the system. We assume an overlapped TRP group (OTG) configuration where some TRPs can belong to multiple adjacent TRP groups. The OTG configuration improves the transmission quality for UEs positioned at the coverage-area border of adjacent TRP groups. The proposed method is actualized through autonomous decentralized control among UEs and TRP groups, which is favorable for implementation in a large-scale network. In the proposed method, each TRP group independently informs all UEs of supplementary information regarding the bandwidth allocated to newly connected UEs. Based on this information, each UE calculates a metric for selecting a TRP group and feeds back the metric value to the best TRP group showing the highest metric. Finally, each TRP group determines the UE to be newly connected, i.e., the one to be handed over from other TRP groups, based on the metrics reported by multiple UEs. Furthermore, to utilize fully the OTG configuration, quasi-optimal bandwidth splitting for the transmission of multiple TRP groups at the overlapping TRP is processed in a decentralized manner for given TRP-group-selection results after the update of TRP group selection. By repeating the above process, TRP group selection for all UEs that maximizes the system throughput is achieved. Computer simulations show the effectiveness of the proposed bandwidth splitting among TRP groups and the TRP group selection method in the OTG configuration compared to the conventional method based on the received signal power.