論文

Wi-Fi channel state information (CSI)-based fall detection systems have a great potential compared with other alternatives since they are nonintrusive and nonspace limited. However, in the conventional work on Wi-Fi CSI-based fall detection, a phenomenon is commonly observed: the classification performance degrades when data in different environments are used for learning and testing. Nonetheless, when the signal-to-noise-power ratio (SNR) is small, the conventional methods cannot capture features of motion and cannot segment signals accurately. Therefore, there is a need to address these problems in order to build a robust fall detection system. In this article, we propose a spectrogram-image-based fall detection using Wi-Fi CSI. Unlike the conventional method, CSI is segmented with a certain sliding-time window, and then the classifier detects fall by using the spectrogram image generated from the segmented CSI. We use a pretrained convolutional neural network (CNN) optimized for binary classification of the spectrogram images of the fall and nonfall motions. We carried out experiments to evaluate the classification performance of our proposed method against the conventional one by using motion data in two different rooms for learning and testing. As a result, we confirmed that our proposed method outperforms the conventional one and reaches over 0.92 accuracy. In addition, compared with the conventional method, the fall detection performance of our method does not degrade even when using different environment data for learning and testing.

Nowadays, microgrids (MG) have attracted much attention, as a key technology of the Internet of Energy (IoE). A great deal of research have shown that the hierarchical microgrid is a more novel structure of IoE. Although the hierarchical microgrid model solves the problem of weak power scheduling capability across microgrids, it suffers from severe communications uncertainty, which can lead to communication delay and fluctuation. To obtain the accurate result of the renewable energy accommodation assessment capacity, a hierarchical microgrid model considering communication uncertainty is proposed in this article. The solution to solve the problem of the assessment renewable energy accommodation capacity for hierarchical MG is a hybrid control based on distribution deep reinforcement learning. The temporal difference (TD) generation adversarial network (TD-GAN) is proposed as a value-based method. Compared with the policy-based method, it can better solve the distributed problem in hybrid control with a generation adversarial network (GAN). Moreover, the challenge that the method cannot handle a continuous action space is solved by using a normalized advantage function (NAF). The method similar with the TD error method is employed to train the GAN network. Simulation results using real power grid data demonstrate the effectiveness and accuracy of the proposed method.

With the rapid development of information technology, the multisource heterogeneous data containing meaningful information have been significantly generated by various edge devices in Internet of Energy, which is one of essential foundations of many knowledge discovery tasks based on edge computing. For some complicated tasks, essential features are owned by different data sellers offering data by blockchains. With limited budgets, buying features are crucial steps in knowledge discovery tasks in electricity spot markets, especially for learning-based algorithms. However, there are lack of proper data pricing mechanisms tailored to dynamic learning processes. Besides, existing methods cannot efficiently employ edge computing servers to obtain optimal policies for selecting features according to dynamic pricing with limited budgets. To overcome such drawbacks, a data pricing mechanism is proposed in this article, which consists of static and dynamic pricing parts. Based on this mechanism, given limited budgets, a feature selection (FS) algorithm considering multiple new factors is proposed, which offers near-optimal solutions for FS at different scenarios. Numeric results show the effectiveness of the proposed algorithms.

In this paper, a channel state information (CSI) feedback method is proposed based on deep transfer learning (DTL). The proposed method addresses the problem of high training cost of downlink CSI feedback network in frequency division duplexing (FDD) massive multiple-input multiple-output (MIMO) systems. In particular, we obtain the models of different wireless channel environments at low training cost by fine-tuning the pre-trained model with a relatively small number of samples. In addition, the effects of different layers on training cost and model performance are discussed. Furthermore, a model-agnostic meta-learning (MAML)-based method is proposed to solve the problem associated with large number of samples of a wireless channel environment required to train a deep neural network (DNN) as a pre-trained model. Our results show that the performance of the DTL-based method is comparable with that of the DNN trained with a large number of samples, which demonstrates the effectiveness and superiority of the proposed method. At the same time, although there is a certain performance loss compared with the DTL-based method, the MAML-based method shows good performance in terms of the normalized mean square error (NMSE).

In the 5th generation mobile communications (5G) and 5G and beyond (B5G), a high altitude platform station (HAPS) is expected to serve as a flying base station (BS) to provide communications over wide areas. In the HAPS system, a multi-cell configuration with multiple beams is considered to increase system throughput. When the HAPS is subjected to wind pressure, the cell range moves accordingly, causing degradation of received signal power and handover to the user equipment (UE). To suppress such degradation and handover, beam control of HAPS is necessary. However, it is not easy to control the beam because multiple antenna parameters affect each other and determine the cell range. In this paper, we propose a beam control method for HAPS using fuzzy Q-learning in multi-cell configuration. In this type of learning, the variable states are controlled by the use of fuzzy sets, which allows multiple searches to be performed in one setup, thus reducing the cost of search, compared with conventional Q-learning. In the proposed beam control method, antenna parameters are controlled by fuzzy Q-learning so that the number of users having a received signal power larger than a predetermined threshold becomes larger in each cell. We evaluate the proposed method by computer simulation and show that the proposed method can improve the number of users having a received signal power larger than a predefined threshold and thus reduce the number of users with low throughput compared to before learning.

In this research, we propose a novel antenna control method for reducing the number of low throughput User Equipments (UEs) caused by the movement and rotation of High-altitude platform station (HAPS). We assume that each HAPS has three antenna arrays for serving one region that consists of three cells and that we know the UE locations in each cell. In the proposed method we first redesign the cell configuration for each HAPS and then divide all UEs into three cells based on the UE locations. Based on the radius and center location information of three new cells, we can mathematically calculate the antenna parameters by the desired coverage model. Thus, each antenna array will be controlled to serve a new cell, respectively. We evaluate the proposed method under 5 different UE distribution scenarios. The simulation results show that, in 5 different UE distribution scenarios, the proposed method can reduce the number of UEs with low throughput. Compared with the conventional method, the proposed method can achieve good throughput performance in all the scenarios.

With the ubiquitous deployment and applications of Internet of Things (IoT), security issues pose a critical challenge to IoT devices. External attackers often utilize vulnerable IoT devices to invade the target's internal network and then further cause a security threat to the whole network. To prevent such attacks, it is necessary to develop a security mechanism to control the access of suspicious IoT devices and manage the internal devices. In recent years, deep learning (DL) algorithm has been widely used in the field of edge device identification (EDI), and has made great achievements. However, these previous methods are essentially centralized learning-based EDI (CentEDI) that trains all data together, which can not guarantee data security and not conducive to deployment on edge devices. To address this problem, we introduce a federated learning-based EDI (FedeEDI) method via network traffic to automatically identify edge devices connected to the whole network. Experimental results show that the training efficiency of our proposed FedeEDI method is much higher than that of the CentEDI method, although its classification accuracy is slightly reduced. In contrast to the CentEDI method, the proposed FedeEDI method has two main advantages: faster training speed and safer training process.

Intelligent reflecting surface (IRS)-aided millimeter-wave (mmWave) multiple-input single-output (MISO) is considered one of the promising techniques in next-generation wireless communication. However, existing beamforming methods for IRS-aided mm Wave MISO systems require high computational power, so it cannot be widely used. In this paper, we combine an unsupervised learning-based fast beamforming method with IRS-aided MISO systems, to significantly reduce the computational complexity of this system. Specifically, a new beamforming design method is proposed by adopting the feature fusion means in unsupervised learning. By designing a specific loss function, the beamforming can be obtained to make the spectrum more efficient, and the complexity is lower than that of the existing algorithms. Simulation results show that the proposed beamforming method can effectively reduce the computational complexity while obtaining relatively good performance results.

In frequency-division duplexing (FDD) massive multiple-input multiple-output (MIMO) systems, deep learning for predicting the downlink channel state information (DL-CSI) has been extensively studied. However, in some small cellular base stations (SBSs), a small amount of training data is insufficient to produce an excellent model for CSI prediction. Traditional centralized learning (CL) based method brings all the data together for training, which can lead to overwhelming communication overheads. In this work, we introduce a federated learning (FL) based framework for DL-CSI prediction, where the global model is trained at the macro base station (MBS) by collecting the local models from the edge SBSs. We propose a novel model aggregation algorithm, which updates the global model twice by considering the local model weights and the local gradients, respectively. Numerical simulations show that the proposed aggregation algorithm can make the global model converge faster and perform better than the compared schemes. The performance of the FL architecture is close to that of the CL-based method, and the transmission overheads are much fewer.

Arrhythmia is one of the causes of sudden infant death, and it is very important to detect fetal arrhythmia for fetal well-being. Fetal electrocardiogram (FECG) is one of the methods to detect a heartbeat. Fetal arrhythmia can be detected based on the heartbeat detection results from FECG signals such as heartbeat intervals. However, the accuracy of arrhythmia detection easily degrades depending on the accuracy of heartbeat detection. In this paper, we propose a deep learning-based fetal arrhythmia detection method using FECG signals. Recently, arrhythmia detection methods using adult ECG signals have achieved a high arrhythmia detection accuracy based on deep learning. Motivated by this fact, in the proposed method, the acquired FECG signals are segmented, and the segments are input into a deep learning model that classifies them into normal or arrhythmia ones. Based on the classification results of multiple segments, a subject is judged as a healthy or arrhythmia subject. Each segment of the training data is divided into three categories based on the estimated heartbeat interval: (i) normal, (ii) arrhythmia, and (iii) a segment that could be both normal and arrhythmic. Only segments labeled as normal or arrhythmia are used for training a deep learning model to achieve a higher classification accuracy of the model. Through these procedures, the proposed method detects fetal arrhythmia with fewer effects of heartbeat detection results. The experimental results show that the proposed method achieves 96.2% accuracy, 100% specificity, and 100% recall, improving the values of conventional methods based on heartbeat detection and feature detection.

The vision and development of blind source separation (BSS) considering the multi-dimensional components attract growing interests in the research community. The single-set BSS methods cannot address the challenge of aligning the extracted components across different datasets, because they neglect the dependence information across datasets. In this paper, we propose to apply the tensor theory to develop a novel structure for the BSS technique that allows us to discover the multi-set components and meaningful spatio-temporal structures of data. The structure is stacked of covariance matrices with temporal information, in terms of limited storage, using the tensor theory that can efficiently encapsulate and compress large-scale data into a compact format. In addition, we discuss how our results provide unique decomposition under such conditions from the theoretical perspective. The experiments designed on the signal with time variation (multiset), which is not identifiable when each mixture is considered individually. It can be solved to restate as a tensor structure of their spatial and temporal correlation matrices. Compared with the other three classical algorithms, the proposed algorithm consistently shows high accuracy.

Due to the lack of channel reciprocity in frequency division duplexity (FDD) massive multiple-input multiple-output (MIMO) systems, it is impossible to infer the downlink channel state information (CSI) directly from its reciprocal uplink CSI. Hence, the estimated downlink CSI needs to be continuously fed back to the base station (BS) from the user equipment (UE), consuming valuable bandwidth resources. This is exacerbated, in massive MIMO, with the increase of the antennas at the BS. This paper propose a fully convolutional neural network (FullyConv) to compress and decompress the downlink CSI. FullyConv will improve the reconstruction accuracy of downlink CSI and reduce the training parameters and computational resources. Besides, we add a quantization module in the encoder and a dequantization module in the decoder of the FullyConv to simulate a real feedback scenario. Experimental results demonstrate that the proposed FullyConv is better than the baseline on reconstruction performance and reduction of the storage and computational overhead. Furthermore, the FullyConv added quantization and dequantization modules is robust to quantization error in real feedback scenarios.

Internet of energy (IoE) is one of the most complex industrial systems, and its stable operation is very important. Situation awareness (SA) has been proposed to ensure the stable operation for IoE and making full use of the relationships between components has become the key point for designing an efficient SA model. In this article, graph digital twin (GDT) is proposed by combining digital twin technology with graph theory, to describe the logical relationships between physical entities more accurately in digital space, and then a novel SA model for IoE based on GDT is proposed. In order to make full use of the relationship between nodes, two classifiers based on graph convolution network are designed for fault location and stability prediction. The experimental results show that the proposed SA model can localize the multiple fault components with high accuracy, and can accurately predict the stability of the system.

Electrocardiogram (ECG) is a common used noninvasive test to quickly detect the heart problem with high precision. However, devices used to collect continuous ECG waveforms under free-living conditions have several operational difficulties. As an alternative, photoplethysmography (PPG) signals can be conveniently collected by pulse sensors, which can be mounted onto wearable devices. In this paper, we study the relation between ECG waveforms and PPG signals by proposing a novel growing hybrid polynomial network (GHPN)). Based on the projections between adjacent layers, the network is designed to approximate the target signals with the recorded inputs and the multi-scale dynamics of the input series are explored gradually through the growth of layers. Two public datasets are employed to evaluate the performance of the proposed approach on the accuracy of waveform reconstruction and heart rate (HR) detection with the widely used metrics. Compared with the reference ECG waveforms, the normalized mean square error (NMSE) of the proposed approach is 0.248 and 0.216 for PPG-Dalia and CapnoBase datasets, which is smaller than the comparison approaches. The average absolute value (AAE) between the detected HR and reference HR is 0.93 and 1.05 for PPG-Dalia and CapnoBase datasets, which exhibit higher HR detection accuracy. Experimental results obtained from benchmark datasets clearly show that the proposed method can achieve higher similarity on the reconstructed morphology with higher HR detection accuracy. Moreover, the proposed approach make it possible to employ PPG sensors for long-term monitoring of the heart actions with higher precision.

Non-contact measurement of the heart rate (HR) refers to the usage of wireless sensors (e.g., Doppler sensors) to identify the heart-related faint movements and to reconstruct the R-peaks or estimate the HR. Conventional work in the field requires the chest of the subject being monitored to be in a specific position in front of the sensor. In our previous work, we have proposed to use a Multiple-Input Multiple-Output (MIMO) Doppler Radar to perform the detection of heartbeats even when the Signal-to-Noise Ratio (SNR) is not very high. By exploiting the fact that different beams have different SNR of heartbeat components, our previous method detects heartbeats by aggregating several received signals. In general, it is essential to use beam directions toward the a subject's chest. However, the chest location estimation is challenging, when there exist several subjects or other moving objects. In the current paper, to deal with this issue, we narrow down the direction of beams to use by using a system composed of a MIMO Doppler sensor and a Light Detection and Ranging (LiDAR) device. By means of transfer learning and automatic annotation of data, the LiDAR is trained to identify the relative chest angle and distance to the subject from the devices, while the MIMO Doppler Radar can adjust the beam towards the identified direction. Throughout experiments, we show the effectiveness of the combination of both the Lidar and radar in detecting the direction and distance to the subject as well as the heartbeat.

Convolutional neural network (CNN) is one of the most important tools to accomplish high-spatial-resolution remote sensing (HSRRS) image classification tasks with their unique feature extraction and feature expression capabilities. However, the CNN-based classification method is very limited due to the acquisition of HSRRS images is difficult and the sample size is limited. In addition, the extraction of features by a single model is very limited, which limits the further improvement of classification performance. To solve the above problems, we propose ResNet50-InceptionV3 based on deep transfer learning and multi-feature fusion (TLMFFRI) model to apply for high-spatial-resolution remote sensing image classification. First, both ResNet50 and InceptionV3 are trained on the ImageNet dataset. Then, transfer the trained convolutional layers weights to the TLMFFRI model to fuse the features and realize the HSRRS image classification. Finally, we evaluate the method on the HSRRS dataset. Compared with ResNet50 based on transfer learning (TL-ResNet50) and InceptionV3 based on transfer learning (TL-InceptionV3), the proposed method achieved better classification performance.

This paper discusses a cellular network assisted by an energy-and spectral-efficient unmanned aerial vehicle (UAV), in which the UAV is deployed to serve mobile users in the cellular network and enable mobile data offloading from a ground base station (GBS) by taking a circular flight route. We explore a visible light communication (VLC)-enabled UAV, in which a light-emitting diode (LED) is mounted on a rotary-wing UAV to offer communications to the users. Our aim is to simultaneously optimize both energy efficiency (EE) and spectral efficiency (SE) of the VLC-enabled UAV by jointly optimizing the common throughput of all users as well as the UAV’s trajectory and flying speed. We employ a unified metric, called resource efficiency (RE), and explore the RE optimization to obtain an adaptive EE-SE tradeoff. The problem posed is seen in a complex and non-convex shape, making it hard to solve. Motivated by the enormous achievement of deep reinforcement learning (DRL) in solving complex control problems, we propose a DRL-based approach to handle this non-convex and complicated optimization. The findings of the simulation reveal that the developed framework achieves a substantial performance in terms of the solution convergence as well as the promising quality of the solutions.

Cuff-less blood pressure (BP) estimation methods using a deep learning model from photoplethysmogram (PPG) or electrocardiogram (ECG) have been actively studied in recent years. However, we found that most previous studies incur data leakage, where segments or records measured from the same subject appear in both the training and test datasets. Furthermore, many previous studies are suspected to have misinterpreted a record in the public dataset used for their evaluations as a subject. To investigate data leakage in BP estimation methods, this paper first organizes previous studies in terms of data leakage. We then quantitatively evaluate the effect of data leakage caused by the segment-level and the record-level train-test split using the public dataset, Cuff-Less Blood Pressure Estimation Data Set. Our experimental results showed that the segment-level split and record-level split erroneously improved the estimation accuracy of mean blood pressure from the (quasi-)subject-level split by the Pearson’s correlation coefficient of 0.56 and 0.40 when using PPG, and 0.82 and 0.69 when using ECG, respectively. These results confirmed that the train-test split used in many previous studies, including the one that described its evaluation as causing no data leakage, causes a high level of data leakage, and that a record in the Cuff-Less Blood Pressure Estimation Data Set, often misinterpreted as a subject in previous studies, is not a subject and that a high level of data leakage occurs when the record is considered as a subject.

This paper proposes an unmanned aerial vehicle (UAV)-aided full-duplex non-orthogonal multiple access (FD-NOMA) method to improve spectrum efficiency. Here, UAV is utilized to partially relay uplink data and achieve channel differentiation. Successive interference cancellation algorithm is used to eliminate the interference from different directions in FD-NOMA systems. Firstly, a joint optimization problem is formulated for the uplink and downlink resource allocation of transceivers and UAV relay. The receiver determination is performed using an access-priority method. Based on the results of the receiver determination, the initial power of ground users (GUs), UAV, and base station is calculated. According to the minimum sum of the uplink transmission power, the Hungarian algorithm is utilized to pair the users. Secondly, the subchannels are assigned to the paired GUs and the UAV by a message-passing algorithm. Finally, the transmission power of the GUs and the UAV is jointly fine-tuned using the proposed access control methods. Simulation results confirm that the proposed method achieves higher performance than state-of-the-art orthogonal frequency division multiple-access method in terms of spectrum efficiency, energy efficiency, and access ratio of the ground users.

Matrix completion-assisted channel estimation is considered one of promising techniques in millimeter wave (mmWave) massive multiple input multiple output (MIMO) system by exploiting the low-rank property of channel matrix in the angle domain. However, existing channel estimation approaches are hard to achieve high accuracy due to the inevitable bias solution caused by nuclear norm based minimization (NNM). To address this problem, this paper proposes a novel matrix completion-assisted mmWave massive MIMO channel estimation method. We employ an effective and flexible rank function named joint weighted and truncated nuclear norm as relaxation of nuclear norm, and then construct an novel matrix completion model for channel estimation problem. Moreover, a popular framework of alternating direction method of multipliers (ADMM) is derived for minimization of the resulting optimization problem. Simulation results are provided to verify the proposed method that can flexibly and effectively improve the channel estimation accuracy with reliable convergence.