Published papers

End-to-end network slicing in uncrewed aerial vehicle (UAV) networks has attracted significant attention for delivering differentiated services tailored to the quality of service (QoS) requirements of user terminals (UTs). In a sliced UAV network, the dual mobility of both UTs and UAVs, combined with the three-layer association of UT-UAV-network slice (NS), makes handoff a critical and challenging problem for maintaining the communication performance of UTs. This paper proposes a sequential handoff process where a central controller makes handoff decisions for each UT sequentially according to the UT queue. This process is modeled as a Markov decision process (MDP). In the MDP, the state space includes NS available resources, UTs' QoS requirements, and UT queue information. Additionally, the reward function accounts for UTs' QoS and handoff overhead. The double deep Q-network (DDQN) technique is then employed to design a handoff algorithm. This algorithm achieves the optimization goal of ensuring UTs' QoS requirements while avoiding frequent handoffs by maximizing the total reward of the MDP. Additionally, a state-dependent exploration technique with adaptive intensity adjustment, called noisy network (NoisyNet), is utilized to drive the DDQN for effective exploration, thereby enhancing the performance of the handoff algorithm. Finally, the runtime complexity of the proposed NoisyNet-DDQN-based sequential handoff (NoisyNet-DDQN-SH) algorithm is analyzed. The simulation results show that the NoisyNet-DDQN-SH algorithm converges well. Moreover, it exhibits strong robustness. Additionally, the NoisyNet technique effectively enhances the algorithm's performance, making the NoisyNet-DDQN-SH algorithm superior to the vanilla DDQN (with an \varepsilon-greedy strategy), greedy, maximum signal-to-interference-plus-noise ratio (Max-SINR), and random algorithms.

This paper presents a radar-based, non-contact Blood Pressure (BP) estimation model based on accurate detection of cardiac activities, which enables BP monitoring to be performed in a touch-free and continuous manner. Cardiac movements have been regarded as essential factors for BP estimation. However, accurately obtaining these movements by a radar system remains a challenging problem because these movements are too inconspicuous and could be easily hindered by respiration and random noise. In this paper, we propose a method that mainly focuses on cardiac feature extraction in radar-based BP monitoring. First, we employ an integrated-spectrum waveform. It is derived from short-time Fourier transform (STFT) and is capable of recording and preserving minor cardiac activities. Compared with the pulse-wave signals used in previous works, the integrated-spectrum focuses on energy changes introduced by short and high-frequency vibrations. It can eliminate the interference of respiration and random noise, and cardiac contractile movement can be reserved accurately. Second, we propose a cardiac features estimation method in which a hidden semi-Markov model (HSMM) is applied to the integrated-spectrum for feature extraction. Compared to the pulse wave signal, the Root-Mean-Square Error (RMSE) of the estimated interbeat intervals (IBI), Systolic time, and Diastolic time is reduced by 44.2%, 73.3%, and 76.7% respectively. The estimated accurate cardiac features are further used as inputs for a Random Forest model for BP prediction. Though previous work required the subject to hold one's breath, we achieved a comparable prediction accuracy even when our subject is breathing normally. The Diastolic BP (DBP) error of our model is 4.27± 5.84 mmHg (Mean Absolute Difference ± Standard Deviation), and the Systolic BP (SBP) error is 6.63± 8.95 mmHg.

Vital signs such as heart rate are important indicators for assessing human health. In particular, noninvasive heart rate measurement using non-contact devices such as continuous wave (CW) radar has attracted attention as a promising technique for estimating heart rate by detecting small cardiovascular movements. However, existing estimation methods have difficulty in accurately estimating heart rate when strongly affected by noise such as respiration and body movement. In this research, we propose a heart rate estimation method with high robustness against noise by combining a pre-processing method using spectrogram and signal decomposition based on Variational Mode Decomposition (VMD). In the proposed method, preprocessing based on the spectrogram is first performed before applying VMD. The obtained integrated spectrum containing the heartbeat-related components is then processed using VMD, which decomposes it into multiple intrinsic mode function (IMF) signals. To select the IMF signals containing heartbeat components, the autocorrelation function and the frequency variation characteristic of heartbeat signals are used as criteria for selection. To evaluate the characteristics of the proposed method, measurements were performed on 10 subjects in a sitting position while electrocardiogram (ECG) was simultaneously recorded as a reference signal. Experimental results showed that the proposed method achieved a low mean absolute error (MAE) of 5.46 bpm compared to the conventional methods, effectively suppressing the effect of noise and improving the accuracy of heart rate estimation.

Traditional machine learning techniques have achieved great success in improving data-rate performance and reducing latency in millimeter wave (mmWave) communications. However, these methods still face two key challenges: (i) their reliance on large-scale paired data for model training and tuning, which limits performance gains and makes beam predictions outdated, especially in multi-user mmWave systems with large antenna arrays, and (ii) meta-learning (ML)-based beamforming solutions are prone to overfitting when trained on a limited number of tasks. To address these issues, we propose a memristorbased meta-learning (M-ML) framework for predicting mmWave beam in real time. The M-ML framework generates optimal initialization parameters during the training phase, providing a strong starting point for adapting to unknown environments during the testing phase. By leveraging memory to store key data, M-ML ensures the predicted beamforming vectors are wellsuited to episodically dynamic channel distributions, even when testing and training environments do not align. Simulation results show that our approach delivers high prediction accuracy in new environments, without relying on large datasets. Moreover, MML enhances the model's generalization ability and adaptability.

Pose estimation and human action recognition (HAR) are pivotal technologies spanning various domains. While the image-based pose estimation and HAR are widely admired for their superior performance, they lack in privacy protection and suboptimal performance in low-light and dark environments. This paper exploits the capabilities of millimeter-wave (mmWave) radar technology for human pose estimation by processing radar data with Graph Neural Network (GNN) architecture, coupled with the attention mechanism. Our goal is to capture the finer details of the radar point cloud to improve the pose estimation performance. To this end, we present a unique feature extraction technique that exploits the full potential of the GNN processing method for pose estimation. Our model mmGAT demonstrates remarkable performance on two publicly available benchmark mmWave datasets and establishes new state of the art results in most scenarios in terms of human pose estimation. Our approach achieves a noteworthy reduction of pose estimation mean per joint position error (MPJPE) by 35.6% and PA-MPJPE by 14.1% from the current state of the art benchmark within this domain.

Millimeter-wave (mmWave) radar-based pose estimation and activity recognition provide robust, precise tracking of human movement, unaffected by lighting or environmental challenges. However, existing human pose estimation methods using point cloud data typically rely on single-frame inputs or apply voxelization, both of which have inherent limitations. Moreover, the conventional methods for activity recognition using point cloud often rely on multistep processes that lose contextual information and depend on pose accuracy. To address these issues, in this paper, we propose mmPoint-Attention, a unified attention and Transformer-based model that directly utilizes multiframe point cloud data for both pose estimation and activity recognition without voxelization. Leveraging temporal context and attention mechanisms, our model achieves accurate 3-D joint estimation and activity classification, with postprocessing to enhance sequence consistency. This approach enables high recognition accuracy and computational efficiency across diverse activities compared with conventional methods.

This paper proposes a decentralized automatic modulation classification (DecentAMC) method using light network and model aggregation. Specifically, the lightweight network is designed by separable convolution neural network (S-CNN), in which the separable convolution layer is utilized to replace the standard convolution layer and most of the fully connected layers are cut off, the model aggregation is realized by a central device (CD) for edge device (ED) model weights aggregation and multiple EDs for ED model training. Simulation results show that the model complexity of S-CNN is decreased by about 94% while the average CCP is degraded by less than 1% when compared with CNN and that the proposed AMC method improves the training efficiency when compared with the centralized AMC (CentAMC) using S-CNN.

In this paper, we propose a method that uses low-resolution infrared (IR) array sensors to identify the presence and location of people indoors. In the first step, we introduce a method that uses 32 × 24 pixels IR array sensors and relies on deep learning to detect the presence and location of up to three people with an accuracy reaching 97.84%. The approach detects the presence of a single person with an accuracy equal to 100%. In the second step, we use lower end IR array sensors with even lower resolution (16 × 12 and 8 × 6) to perform the same tasks. We invoke super resolution and denoising techniques to faithfully upscale the low-resolution images into higher resolution ones. We then perform classification tasks and identify the number of people and their locations. Our experiments show that it is possible to detect up to three people and a single person with accuracy equal to 94.90 and 99.85%, respectively, when using frames of size 16 × 12. For frames of size 8 × 6, the accuracy reaches 86.79 and 97.59%, respectively. Compared to a much complex network (i.e., RetinaNet), our method presents an improvement of over 8% in detection.

This paper presents a new solar-powered fixed-wing unmanned aerial vehicle (UAV)-assisted data collection technique, where a fixed-wing UAV harvests solar energy to fly and collect data from smart devices (SDs). The minimum of the data upload-ed from any of the SDs is maximized by optimizing the UAV's three-dimensional (3D) trajectory. The key idea is that we develop variable substitution and successive convex approximation (SCA) techniques to convexify the nonconvex data transmission, UAV energy consumption and mobility constraints, as well as the nonconvex energy harvesting constraint. The proposed algorithm guarantees a locally optimal solution satisfying the Karush-Kuhn-Tucker (KKT) conditions. Another important aspect is that an effective set of approximation parameters are identified for accurate approximation of the solar energy harvesting at relatively low altitudes, contributing to the design of efficient 3D trajectories balancing energy harvesting and data collection. Numerical results show that the 3D flight increases uploaded data by 59% as compared to the two-dimensional (2D) flight.

In this paper, we propose a joint optimization algorithm of user pairing, power allocation, and content server (CS) deployment in non-orthogonal multiple access (NOMA) assisted wireless caching networks (WCN). The main purpose of the proposed algorithm is to maximize the hit probability of users, which is determined by the hit probability of CSs and the decoding success probability of users. Since the optimization problem is non-convex and difficult to solve directly, it is decomposed into three sub-problems, user pairing, power allocation, and CS deployment. First, users are served by CSs in a hybrid NOMA manner and paired according to the power demand coefficient. Secondly, we derive the closed-form solution of the power allocation factor in the Rayleigh fading channels. Thirdly, the simulated annealing algorithm is used to optimize the deployment of CSs. Finally, simulation results confirm that the proposed joint optimization scheme outperforms the benchmark scheme.

High Altitude Platform Station (HAPS), functioning as a flying base station (BS) positioned in the stratosphere, has gradually captured more interest in the field of mobile communications. HAPS holds the potential to deliver extensive coverage areas and resilient networks in the face of disasters. Using beamforming techniques, a HAPS can configure multiple cells within its coverage area to serve a large user population. However, the position of HAPS and the distribution of user equipment (UE) changes with time, resulting in changes in the relative position between HAPS and UEs. Consequently, the maximum throughput of UE decreases and some UEs can suffer network outages. A previously proposed method treats this as an optimization problem and tries to maximize throughput based on an objective function. However, the computational complexity is still too high for real-time control. To help beams respond faster to position changes, some AI-based methods are proposed but they cannot guarantee that low throughput UEs are well minimized. In this paper, we propose a K-means clustering-aided particle swarm optimization (PSO) algorithm that can adjust the cell configuration to minimize the number of low throughput UEs. Taking advantage of K-means clustering, this method redefines the search range for each parameter to help PSO quickly converge to the optimum. By running simulations using realistic UE distributions in some cities, we demonstrate the superiority of our proposed method in terms of computational complexity and ability to reduce low throughput UEs.

The rapid advancement of the Internet of Vehicles (IoV) has created significant challenges in large-scale data processing and low-latency decision-making. Multi-access edge computing substantially enhances data processing capabilities and system response times by pushing computational resources to the network edge. Task offloading technology transfers vehicle computing tasks to edge servers, alleviating the computational burden on user devices. Traditional offloading methods struggle with the challenges posed by the rapid growth of IoV. With the advancement of deep learning, large models have emerged as powerful tools for handling complex tasks. In this paper, we propose a task offloading algorithm that integrates a Transformer-based large model with deep reinforcement learning. We utilize the Transformer-based large model to replace the action generation network in deep reinforcement learning. The proposed method utilizes extensive historical data and real-time task information to dynamically adjust task allocation schemes. This approach enables the conclusion of task offloading strategies that minimize energy consumption and latency. Additionally, we propose a task classification framework to further reduce offloading delays and energy consumption. The framework categorizes tasks based on key attributes such as computational complexity, data size, and time sensitivity. We propose a reward function based on task adaptability to conclude the task offloading strategy that minimizes both offloading delay and energy consumption. Experimental results demonstrate that the proposed algorithm can accurately and quickly conclude the task offloading strategy to minimize energy consumption and latency.

As the global population continues to age and technologies, such as the Internet of Things (IoT) and edge computing advance rapidly, indoor human activity recognition (HAR)-based on Wi-Fi channel state information (CSI) has gained significant research attention. However, the high computational complexity of existing HAR methods limits their deployment on resource-constrained devices. To address this challenge, we propose a lightweight HAR method using branch decision lightweight two-stream convolution-augmented transformer (BLTHAT) model, which integrates depthwise separable convolutions (DSC) and an improved framework structure to enhance computational efficiency. Additionally, we introduce the branch fusion network (BFN), a decision-making module designed to optimize feature processing and improve model robustness. Further enhancements in attention mechanisms and regularization strategies contribute to reducing complexity while maintaining high recognition accuracy. Comprehensive experiments were conducted on a multilabel dataset. The results demonstrate that our proposed HAR method achieves high computational efficiency with minimal complexity, making it well-suited for IoT applications. Ablation studies further confirm that the multibranch structure of the BFN module enhances feature extraction without significantly increasing computational overhead.

Human activity recognition (HAR) with Wi-Fi channel state information (CSI) is attractive for privacy-preserving, device-free sensing; yet, real deployments still struggle with three coupled issues: robustness across rooms and bands, efficiency on edge hardware, and unified support for multiple tasks. We present UN-2DCNN, a lightweight 2-D convolutional neural network (CNN) pipeline tailored to indoor, multiuser CSI sensing. The design reduces temporal redundancy via a simple temporal skipping augmentation, learns a compact 128-D representation with a small CNN + global average pooling (GAP) backbone, and injects reliability feedback through uncertainty-aware feature scaling (UAFS): Stage-1 predictive entropy is mapped to a gating weight that rescales features before a second decision head. A channel-attention MLP further suppresses spurious subcarrier responses. Evaluated on a recent multiuser CSI benchmark across classrooms, meeting rooms, and empty environments at 2.4/5 GHz, UN-2DCNN consistently outperforms competitive recurrent neural network (RNN)/Transformer baselines while using only ~1M parameters and maintaining sub-2-s test-time latency. Beyond higher accuracy, the model exhibits faster, smoother convergence and improved calibration (fewer overconfident errors). Ablations confirm that removing attention, UAFS, or the second-stage head yields consistent drops, and simple temporal skipping on the data side complements model-side selectivity. These results indicate that reliability-aware, lightweight designs can deliver practical accuracy–efficiency tradeoffs for CSI-based perception on edge/IoT platforms.

In emerging smart cities and intelligent transportation systems (ITS) ecosystems, sixth-generation (6G) vehicular networks demand sustainable wireless environments capable of adapting to high mobility and dynamic propagation conditions. In dynamic vehicular networks, irregular reconfigurable intelligent surface (IRIS) with non-uniformly distributed elements over a larger surface offer improved coverage, reduced interference, and lower power consumption. This article undertakes a rigorous investigation into IRIS-assisted downlink vehicular communication in the context of Doppler shift, with the objective of optimizing the long-term net weighted average sum-rate under both perfect and imperfect channel state information (CSI). This task entails a comprehensive joint optimization of IRIS topology selection, phase-shift matrix, and beamforming vectors, while ensuring queue stability. To address this stochastic network optimization problems involving time-varying system dynamics, a Lyapunov optimization-based framework is proposed. This framework transforms the stochastic problem into a deterministic form, ensuring queue stability without requiring knowledge of future channel conditions. Subsequently, a joint optimization strategy based on the block coordinate descent (BCD) algorithm is applied to make optimal decisions of IRIS topology selection, phase-shift matrix, and beamforming vectors in each time slot. To validate the effectiveness of the proposed framework, extensive simulations were conducted and benchmarked against two established schemes: one employing conventional IRIS configurations in wireless networks and the other utilizing non-orthogonal multiple access (NOMA) techniques. The results consistently demonstrate that the proposed IRIS-assisted Lyapunov optimization framework significantly outperforms both benchmarks in terms of achievable sum-rate and maintains queue stability.

To leverage antenna diversity effectively in Massive Multiple-Input Multiple-Output (MIMO) systems, Channel State Information (CSI) utilization at the base station (BS) is crucial. In Frequency Division Duplexing (FDD) MIMO setups, downlink CSI is acquired by user equipment (UE) and transmitted back to the BS. However, this feedback overhead diminishes communication throughput, necessitating CSI compression technologies. Methods such as traditional compression or more advanced ones such as Transformer-based neural networks which aim to address this issue suffer from low compression rates or high computational cost. In this context, Reformers emerge as a potential solution as they employ Locality-Sensitive Hashing (LSH) attention, offering a comparable accuracy to Transformers but with reduced computational demands. In this paper, we propose a CSI feedback method using Reformer. Specifically, by replacing the multi-head attention of the CSI feedback method using Transformer with the LSH attention of Reformer, we aimed to achieve similar CSI reconstruction accuracy while reducing computational complexity. Computer simulations confirm that the CSI feedback method using Reformer reduces computational complexity by about 12% while achieving comparable CSI reconstruction accuracy compared to the CSI feedback method using Transformer.

Semantic communication has gained significant attention in modern data-intensive applications due to its efficient information transmission capabilities. While recent advancements in large language models and diffusion models have enabled generative image semantic communication in text-based and multimodal settings, existing approaches struggle with visual fidelity due to the inherent limitations of textual descriptions and the lack of dedicated design in multimodal methods. To address these issues, we propose a compressed sensing-driven multimodal semantic communication framework that transmits saliency-adaptive sparse sampling, edge contours, and textual descriptions of images at the transmitter and reconstructs images at the receiver using guided diffusion models. We innovatively design a low-rank guided diffusion (LRGD) model to adapt to this framework, employing a dynamic smoothing low-rank guidance process in the latent space using low-quality reconstructed images. This approach enables highly accurate restoration of visual details under a low sampling ratio without additional training. Compared to state-of-the-art methods, our framework uses model-independent and interpretable universal semantic representations for communication, significantly enhancing cross-model compatibility without model sharing between the transmitter and receiver. Extensive experiments on multiple datasets demonstrate that the proposed method achieves robust and superior image reconstruction performance across various noisy environments while compressing data to 5.7% of the original image size.

Malware traffic classification (MTC) is a crucial step in network intrusion detection, which is significant for network security and management. With the continuous evolution of malware traffic, traditional MTC methods are difficult to adapt efficiently to new traffic categories, and manually designed neural network structures suffer from performance bottlenecks and low design efficiency. Hence, we propose an enhanced MTC method based on expandable class incremental learning (CIL) with architecture search. The architecture search can automatically design the optimal neural network structure tailored to different network traffic characteristics, avoiding the limitations of manually designing network structures and improving classification performance. Meanwhile, expandable CIL allows the MTC model to gradually learn new traffic categories without forgetting previous knowledge, avoiding the computational overhead and efficiency loss caused by frequent retraining of the model. The experimental results demonstrate that the proposed CIL-MTC approach surpasses advanced incremental learning methods on both the Edge-IIoTset and ISCX VPN-nonVPN datasets, achieving superior classification performance while maintaining lower average trainable parameters and training costs. Especially, it achieves an average incremental accuracy of 98.55% and 99.09% on the Edge-IIoTset dataset with incremental tasks of 5 and 2, respectively.

Belief propagation (BP) detection is a technique for separating and detecting incoming signals with minimal complexity in massive multiple-input multiple-output (MIMO) systems. However, because of the interference and noise that remain in the received signals even after attempting to remove them through conventional techniques, errors manifest in the transmitted messages. Due to the MIMO channel's numerous loops, a message containing mistakes spreads across the factor graph leading to a degradation in the BP's convergence properties and detection performance. In this paper, we propose a BP detection using deep image prior (DIP) with deep neural network (DNN)-trained scaling factor. By applying DIP to the BP detection algorithm, we achieve a reduction in residual interference and noise. Post DIP application, there is a modification in the variance of both interference and noise components. To align it more accurately with its true value and enhance message reliability, we adjust the variance using scaling factors trained through DNN-based damped BP (DNN-dBP). Using computer simulations, we demonstrate that applying DIP helps decrease the power of the residual interference and noise after removing the interference at each iteration in the BP detection. It is also shown that the proposed method improves the detection performance compared to the normal BP detection, BP detection without DIP when training the scaling factors of the variance, and BP detection to which DIP is applied without training the scaling factors of the variance.

In Low-Earth Orbit (LEO)-based Non-Terrestrial Networks (NTN), stable and continuous connectivity through efficient handover (HO) mechanisms should be ensured, especially given the high mobility and dynamic satellite coverage in these networks. Conventional HO optimization approaches, such as the Service Continuity Dynamic Programming (SCDP) strategy, optimize HO but may leave some User Equipments (UEs) experiencing low link rates, reduced service time, or frequent handovers. To address this, we propose the Service Continuity Max-Min (SCMM) strategy, which employs the MaxMin algorithm to reduce the number of UEs that have poor connectivity experiences, using a uniquely defined reward function that incorporates link rate, service availability time, and HO frequency. The simulation results demonstrate that SCMM improves the minimum reward across UEs while maintaining an overall performance comparable to that of the existing SCDP methods. This approach offers a robust solution to improve service continuity and user satisfaction in NTN, making it an effective strategy for next-generation satellite networks.