Published papers

© 2013 IEEE. Demands for vital sign monitoring are increasing in the field of health care. In particular, the R-R Interval (RRI) estimation has been studied extensively, since the RRI variation is highly related with the stress of a subject. Various Doppler sensor-based-heartbeat detection methods have been proposed so far, thanks to non-contact and non-invasive features of a Doppler sensor. In our previous research, we have proposed a Doppler sensor-based RRI estimation method by a spectrogram. In this method, the spectra due to heartbeats are integrated on a spectrogram, and then the RRI is estimated by detecting the peaks of the integrated spectrum. However, the undesired peaks sometimes appear even in the situation where a subject sits still. In this paper, as the extended version of our previous method, we propose a Doppler sensor-based RRI estimation method leveraging the accurate peak detection. In the proposed method, to prevent the incorrect peak detection, the peaks due to heartbeats are detected using some peaks before and after the investigated peak. Through the experiments on 10 subjects in the cases where a subject was sitting still, typing, and speaking, we confirmed that the proposed method improved our previous and state-of-the-art ones by the root-mean-squared error of the RRI. Furthermore, based on the estimated RRI, we calculated the stress index low-frequency/high-frequency (LF/HF), which is one of the useful indices to evaluate the stress of a subject. As a result, our proposed method outperformed the other ones by the relative error of the LF/HF.

© 2018 IEEE. With the rapid growth of online social media content, and the impact these have made on people's behavior, many researchers have been interested in studying these media platforms. A major part of their work focused on sentiment analysis and opinion mining. These refer to the automatic identification of opinions of people toward specific topics by analyzing their posts and publications. Multi-class sentiment analysis, in particular, addresses the identification of the exact sentiment conveyed by the user rather than the overall sentiment polarity of his text message or post. That being the case, we introduce a task different from the conventional multi-class classification, which we run on a data set collected from Twitter. We refer to this task as 'quantification.' By the term 'quantification,' we mean the identification of all the existing sentiments within an online post (i.e., tweet) instead of attributing a single sentiment label to it. For this sake, we propose an approach that automatically attributes different scores to each sentiment in a tweet, and selects the sentiments with the highest scores which we judge as conveyed in the text. To reach this target, we added to our previously introduced tool SENTA the necessary components to run and perform such a task. Throughout this work, we present the added components; we study the feasibility of quantification, and propose an approach to perform it on a data set made of tweets for 11 different sentiment classes. The data set was manually labeled and the results of the automatic analysis were checked against the human annotation. Our experiments show the feasibility of this task and reach an F1 score equal to 45.9%.

© 2013 IEEE. A key point for the recovery of a block-sparse signal is how to treat the different sparsity distributed on the different parts of the considered signal. It has been shown recently that grouping the signal, i.e., partitioning the original signal into different groups or segments, and conducting the recovery for these groups separately provides an effective method to deal with the block-structured sparsity and can generate much better performance than the conventional sparse signal recovery (SSR) algorithms. In order to further improve the recovery performance, instead of the fixed grouping method used in the recent results, a novel dynamic grouping method will be first proposed in this paper, which classifies the segments due to the different levels of sparsity in a dynamic way. Then, by incorporating this technique into the block version of adaptive SSR algorithms. we developed recently, i.e., the block zero-attracting least-mean-square (BZA-LMS) algorithm and the block ℓ0-norm LMS (B ℓ0-LMS) algorithm, the corresponding new algorithms, i.e., the BZA-LMS-D and B ℓ0-LMS-D algorithms, will be established. The performance superiorities and the robustness against different block-sparsity and/or noise interference for the new algorithms based on dynamic grouping will be demonstrated by both analytic discussions and numerical simulations for a variety of scenarios.

© 2013 IEEE. Nonlinear blind source separation (BSS) is one of the unsolved problems in unsupervised learning, because the solutions are highly non-unique when there is no prior information for the mixing functions. In this paper, we present a novel approach to tackle the ill-posedness of the nonlinear BSS problem with a few assumptions. The derivation of our algorithm is inspired by the idea of an efficient layer-by-layer representation to approximate the nonlinearity. Once such a representation is built, a final output layer is constructed by solving a convex optimization problem. Relying on the multi-layer architecture, the algorithm transforms a time-invariant nonlinear BSS to the local linear problem with a tolerable computational cost. Then, the projected data can break the nonlinear problem down into the version of a generalized joint diagonalization problem in the feature space. Importantly, the parameters and forms of polynomials depend solely on the input data, which guarantee the robustness of the structure. We thus address the general problem without being restricted to any specific mixture or parametric model. Experiments show that the proposed algorithm has a higher estimation accuracy on audio data sets from the real world for separating various nonlinear mixtures.

© 2019 IEEE. Massive multiple-input multiple-output (MIMO) is a technology that uses many antennas at base station (BS), and can realize advanced beamforming and spatial multiplexing. However, since Massive MIMO has many antennas, there exist problems such as the increase of overhead including pilot signals used for channel estimation and feedback of channel state information (CSI). To reduce the overhead, there is a technique of transmitting pilot signals only from some antennas and obtaining channels of the remaining antennas by the minimum mean square error (MMSE) interpolation based on spatial correlation. In addition, to perform the MMSE interpolation with low computational complexity, there is a method of interpolating channels by dividing the uniform rectangular antenna array (URA) into the vertical and horizontal linear directions. However, there is a problem that channel interpolation accuracy becomes low with dividing into the linear arrays. In this paper, we propose a method of channel interpolation with dividing the entire URA into a set of smaller URAs where interpolation is performed in each URA. Through computer simulation we show that the proposed method reduces the computational complexity compared with that without dividing, and improves the channel interpolation accuracy compared with that with dividing into linear arrays.

© 2013 IEEE. Massive multi-user multiple-input multiple-output (massive MU-MIMO) technology is considered as a promising enabler to fulfill the rapid growth of traffic requirement for wireless mobile communications. The massive MU-MIMO system can achieve unlimited capacity when the base station (BS) has accurate channel state information (CSI). In time-division-duplex (TDD) mode, the BS estimates CSI by receiving pilot signals sent from user terminals (UEs). However, because of using non-orthogonal pilots, pilot contamination happens to degrade the quality of the CSI estimation. To deal with pilot contamination problem, a low-complexity subspace minimum mean square error (MMSE) estimation method is proposed in this paper. Specifically, our approach operates the MMSE estimation in a low-dimensional subspace to avoid large matrix manipulation. Meanwhile, subspace projection helps to discriminate the desired signal and interfering signals in the power domain. Interference analysis shows the MMSE estimation can achieve interference-free estimation even in a low-dimensional subspace with a large number of BS antennas, and non-overlapping angles of arrival (AoAs) between desired and interfering UEs. Furthermore, thanks to the low-rank property of the channel covariance matrix in massive MU-MIMO systems, a two-step covariance matrix subspace projection method is proposed for further computational complexity reduction. The complexity analysis and simulation results indicate that our proposed approach has better channel estimation accuracy with lower complexity than the conventional MMSE estimation when the number of BS antennas is large.

© 2018 IEEE. In recent years, Bitcoin has been used for many services and purposes, e.g. gambling, marketplace, but also even as an investment scam. In order to clarify how Bitcoin is used, it is in great importance to identify what kind of services are operated by Bitcoin addresses. In this paper, we propose a multiclass service identification scheme in Bitcoin based on novel transaction history summarization. Our novelty is to propose how transaction history is retrieved and how the retrieved transactions are processed for better identification. When a Bitcoin address is given, the characteristics of its transaction history is calculated as features. Then, the set of calculated features is fed into a supervised classifier and the services operated by the given Bitcoin addresses are identified among seven major services: (i) exchange, (ii) faucet, (iii) gambling, (iv) investment scam, (v) marketplace, (vi) mining pool, and (vii) mixer. To our knowledge, we are the first to propose a multi-class identification. We show that our scheme achieves 72 % of accuracy through performance evaluation with more than 26,000 Bitcoin addresses that have been used for seven services/purposes from Jan. 2009 to Feb. 2017.

© 2020 The author(s). Sentiment analysis refers to the automatic collection, aggregation, and classification of data collected online into different emotion classes. While most of the work related to sentiment analysis of texts focuses on the binary and ternary classification of these data, the task of multi-class classification has received less attention. Multi-class classification has always been a challenging task given the complexity of natural languages and the difficulty of understanding and mathematically "quantifying"how humans express their feelings. In this paper, we study the task of multi-class classification of online posts of Twitter users, and show how far it is possible to go with the classification, and the limitations and difficulties of this task. The proposed approach of multi-class classification achieves an accuracy of 60.2% for 7 different sentiment classes which, compared to an accuracy of 81.3% for binary classification, emphasizes the effect of having multiple classes on the classification performance. Nonetheless, we propose a novel model to represent the different sentiments and show how this model helps to understand how sentiments are related. The model is then used to analyze the challenges that multi-class classification presents and to highlight possible future enhancements to multi-class classification accuracy.

© 2020 IEEE. Spatia1 modulation (SM) has attracted enormous research attention due to its notable merits in terms of enabling higher spectral- and energy-efficiencies relative to the conventional MIMO technique. To provide higher system performance for SM, linear precoder is designed to shape the APM constellations observed at the receiver. However, numerical results reveal that: (1) The performance of precoding based approaches degrades dramatically for higher-level QAM constellations, which implicitly affect the spectral-efficiency improvement of SM. (2) The complexity of precoder optimization increases greatly with the system dimensions and the APM constellation dimensions, which indeed is cost inefficient. In this context, this paper introduces the signature spatial modulation (SSM) technique that inherits the properties of both the SM and the analog shift weighting techniques. To guarantee superior trade-offs among spectral-and cost-efficiencies, and bit error rate (BER) performance for SSM, we first develop efficient multi-configuration selection mechanisms by exploiting the spatial degree of freedoms (DoFs), as well as by leveraging the benefits of SM. To provide a global optimum solution against the constant-modulus constrained analog precoding problem, we next relax the precoding problem into unconstrained alternating minimization subproblems, followed by proposing Broyden-Fletcher-Goldfarb-Shanno (BFGS) aided alternating minimization algorithm to solve those subproblems. It should be noted that our proposed mechanism can realize an L-fold reduction in terms of the number of transmit antennas. In particular, theoretical results reveal that our proposal can reduce the complexities brought by the configuration selection operation up to 50% when considering L-QAM with L=4.

© 2013 IEEE. In cognitive radio networks (CRNs), if two unlicensed secondary users (SUs) want to communicate with each other, they need to rendezvous with each other on the same channel at the same time. Rendezvous is the first key step for SUs to be able to communicate with each other. Channel hopping (CH) is a representative technique to solve the rendezvous problem in CRNs. SUs equipped with multiple radios can significantly reduce the time-to-rendezvous (TTR) for several existing CH algorithms while the additional cost is low. However, several drawbacks exist in the existing CH algorithms based on multiple radios. One of the main drawbacks is that they cannot be well applied in heterogeneous CRNs. The reason is that the number of radios for different SUs is implicitly assumed same or must be more than one in the existing CH algorithms based on multiple radios, which is unrealistic for heterogeneous CRNs. In heterogeneous CRNs, SUs may be equipped with different numbers of radios including one radio. To mainly address the above issue, hybrid radios rendezvous (HRR) algorithm is proposed in this paper. Moreover, the upper bounds of maximum TTR (MTTR) for the HRR algorithm are derived by a theoretical analysis. Furthermore, extensive simulations are performed to evaluate the expected TTR (ETTR), the MTTR, and the channel qualities of the rendezvous channels for the HRR algorithm. Simulation results show that rendezvous can be guaranteed by the HRR algorithm in heterogenous CRNs. Besides, the qualities of the rendezvous channels can be improved by the HRR algorithm. In addition, our algorithms can achieve rendezvous faster than several existing algorithms.

© 2019 IEEE. Continuous HR (Heart Rate) monitoring enables the stress estimation in daily life. A Doppler sensor could be a key device to facilitate the non-contact HR estimation. As one of the Doppler sensor-based HR estimation methods, we have previously proposed a MUSIC (MUltiple SIgnal Classification)-based HR estimation method. MUSIC is the algorithm widely used as a tool to estimate DOA (Direction of Arrival). In our previous method, MUSIC spectrum is calculated in each sliding window, and then HR is estimated by the maximum peak detection over the MUSIC spectrum. However, when HR changes largely within the window, several peaks due to heartbeats appear over the MUSIC spectrum, which might cause the incorrect peak detection. Hence, an adaptive window is required so that only one peak appears. In this paper, we propose a MUSIC-based HR estimation method with an adaptive window size setting. When several peaks due to heartbeats appear over the MUSIC spectrum, our proposed method shortens the time window and re-calculates the MUSIC spectrum, which is repeated until only one peak appears. The experimental results showed that our method outperformed not only our previous one but also the other existing MUSIC-based HR estimation one in terms of the estimation accuracy of the HR, the stress indexes CVI (Cardiac Vagal Index) and CSI (Cardiac Sympathetic Index).

© 2013 IEEE. Heartbeat detection is one of key techniques to monitor our health condition in daily life, and demands for this technique have increased year and year. Thanks to the non-contact and non-invasive features, various Doppler sensor-based detection methods have been investigated so far. However, the heartbeat detection accuracy of the conventional methods could get degraded due to the low SNR (Signal-to-Noise Ratio) of heartbeat components. Thus, even after some signal processing, non-heartbeat components still remain over such processed signal, which could degrade the heartbeat detection accuracy. In particular for the subjects with low HR (Heart Rate), the estimated HR tends to be higher than the ground truth HR due to such non-heartbeat components, though the conventional methods have mainly focused on the heartbeat detection against the subjects with the normal HR higher than 50 bpm (Beats Per Minute). In this paper, to accurately detect heartbeat even with low HR via a Doppler sensor, we propose a heartbeat detection method based on heartbeat signal reconstruction with convolutional LSTM (Bidirectional-Long Short-Term Memory). In the proposed method, to reconstruct a heartbeat signal based on the periodicity of heartbeat and the spectrum distribution peculiar to heartbeat, successive spectrograms that might be due to heartbeat is used as an input to convolutional LSTM. In addition, for better reconstruction of a heartbeat signal, the previously estimated RRI (R-R Interval) is also used as a feature in the proposed deep learning model with convolutional LSTM. Through the experiments, we confirmed that our proposed method accurately detected heartbeat against 17 subjects including the ones with the HR lower than 50 bpm.

© 2018 IEEE. Heartbeat detection are receiving a lot of attention in the field of health care, since cardiac activity reflects various information of a subject, e.g., the stress. Many Doppler sensor-based heartbeat detection methods have been proposed so far. As one of such methods, the MUSIC (MUltiple SIgnal Classification)-based HR (Heart Rate) estimation method has been proposed. However, the conventional MUSIC-based HR estimation method not only needs a long time window, but also requires to know the number of sinusoidal signals composing the analyzed signals, P, in advance of applying MUSIC to the analyzed signal, which is challenge. In this paper, we propose a MUSIC-based HR estimation method with the DCT (Discrete Cosine Transform)-based parameter P selection. In the proposed method, the analyzed signal is firstly decomposed by DCT. The inverse DCT is then performed based on only components that might be related with heartbeats. The number of components used in the inverse DCT is selected as P. Through the experiments, we confirmed that our method outperformed the conventional one by the estimation accuracy of the HR and the stress indexes such as CVI (Cardiac Vagal Index) and CSI (Cardiac Sympathetic Index).

© 2019 IEEE. Recently, through exploiting the spectral sparsity of heartbeat component, a heartbeat detection method using a stochastic gradient approach has enabled a high-resolution of heartbeat spectrum reconstruction by Doppler radar signal, which also suppresses the residual noises after signal decomposition. However, the interference from respiration and/or body motion often corrupts the decomposition of signal by singular spectrum analysis (SSA), resulting in an inaccurate extraction of heartbeat component. In this paper, a non-negative matrix factorization (NMF)-based blind source separation (BSS) is first applied to non-contact heartbeat detection for better heartbeat extraction, incorporating the stochastic gradient approach. Specifically, motion noise is taken into account as one of sources, achieving relatively stable separation in various scenarios. In our proposed BSS approach, the spectrogram originated from radar signal is decomposed twice by NMF, which is used to learn the basis spectra (BS) relying on spectral correlation. Experimental results showed the improved accuracy and robustness of our method over conventional methods, on the heart rate (HR) measurement against subjects' sitting still or typewriting.

© 2019 IEEE. Heart rate variability is one of major physiological parameters to reflect our stress, which has motivated researchers to investigate a Doppler sensor-based non-contact heartbeat interval estimation algorithm. As one of such methods, we have previously proposed a spectrogram-based method. In this method, the spectrum that might be due to heartbeats is integrated over a spectrogram, and then heartbeat interval is estimated by detecting peaks over the integrated spectrum. However, when a subject moves, undesired peaks with large amplitude appear, which causes the incorrect peak detection. As one of the technique to eliminate the undesired peaks with large amplitude, there is CA-CFAR (Cell Average-Constant False Alarm Rate). CA-CFAR is the technique to detect a signal, when the amplitude of a signal exceeds a threshold calculated with average amplitude of signals before and after the investigated one. However, depending on the duration of body movements, the influence of body movements might be included within the signals used for the threshold calculation, which might results in the detection failure of undesired peaks. This is because the length of GT (Guard Time) is fixed, where GT is the time to prevent the signal used for the threshold calculation from including the investigated signal components. To solve this problem, we propose a novel CA-CFAR in which the length of GT is set as the latest peak interval and only the signal before the investigated one is used so that the influence of body movements does not affect the threshold calculation. Through the experiments where a subject moves, i.e., typing, we confirmed that our spectrogram-based heart rate variability estimation method with the proposed CA-CFAR outperformed the one with CA-CFAR based on fixed GT by the RMSE (Root Mean Squared Error) between the estimated heartbeat interval and the ground truth value of the one.

© 1997-2012 IEEE. In this letter, we study the secrecy performance of the classic Wyner's wiretap model, where the main and eavesdropper channels are modeled by a general and versatile α-η-κ-μ fading model. Novel and exact expressions of the average secrecy capacity and secrecy outage probability have been derived. Previous results on physical layer security can be obtained through our newly derived expressions by specializing the model parameters. More importantly, the derived results are also applicable for the secrecy performance analysis of some field measurements (e.g., in millimeter wave communications), which cannot be analyzed by previous results.

© 2020 IEEE. Identifying objects with radar is in great demand to avoid road accidents. Recent research tried to identify moving objects on a road by inputting radar information to a machine learning classifier. In the conventional method, the features used in the machine learning are extracted from the observed radar information with a short time interval. Since the movement of the objects is different depending on the objects, time series information is effective for classification, which has not been exploited before. In this paper, we propose an on-road object identification considering time series of radar information. We measured objects with 79.5 GHz millimeter-wave radar and extract features from a series of time windows by calculating the mean and variance of object information, i.e., velocity, distance, and signal power. The classification performance was evaluated with a dataset obtained by on-road experiments. It is shown that our method outperforms the conventional one and the proposed features significantly contribute to the accurate identification.

© 2020 IEEE. Electroencephalogram (EEG) signals are important to study the activities of human brains. The independent component analysis (ICA) algorithm is a practical blind source separation (BSS) technique that can separate EEG sources from artifacts effectively. However, most traditional ICA algorithms assume that the mixing process is instantaneous and off-line. In this paper, a novel framework based on the extension of the online recursive ICA algorithm (ORICA) is proposed to apply for motor imagery (MI) EEG recording. The contributions are as follows. Firstly, we show ORICA's adaptability to accurate and effective source separation used for artifact-contaminated MI EEG recording. Secondly, to identify EOG signals on the output of source separation, the topographic map is presented to distinguish the target signals. The experimental results show that the proposed framework is able to be applied to process MI EEG recording in real-time situations.

© 2019 IEEE. The presence of many battery-powered sensors in the Internet of Things (IoT) paradigm calls for the design of green and energy-aware protocols. One of the major challenges in the wireless networks based on IoT is to reduce energy consumption and increase network lifetime. This paper addresses the problem of designing a smart sleep/wake-up mechanism using the genetic algorithm (GA). GA, as a nature-inspired optimization approach, is used to manage the operation modes of the base stations (BSs) according to the time-variant profile of the daily traffic load. Accordingly, when a number of stations are sending data to nodes, i.e., active or awake nodes, other nodes are in inactive (sleep) mode. In this case, the energy efficiency (EE) of the network increases to an acceptable level via controlling the operation modes of the densely deployed BSs over wireless sensor networks (WSNs).

© 2012 IEEE. In this letter, we propose an orthogonality structure design (OSD) for the generalized precoding aided spatial modulation (GPSM) to overcome the performance degradation caused by correlated fading channels. To improve the bit error rate (BER) performance of GPSM system, we devise an efficient receive antenna subset selection (RAS) technique relied on the mechanisms of maximizing the equivalent channel gain. To facilitate a better trade-off between performance and computational complexity, we study the peculiarities of the Hermitian matrix which provides an important insight for conceiving orthogonality conditions to the channel matrix of GPSM system. We prove that the performance gain in terms of BER can be attained with the aids of the proposed efficient RAS, and OSD-aided schemes with low-complexity. In particular, our results demonstrate that the OSD-aided scheme is capable of providing significant BER performance improvement for GPSM systems, compared to that of current strategies.