Singular spectrum analysis (SSA) is a nonparametric spectral estimation method that decomposes time series signals into interpretable components. With the rise of big time series, the demand for effective and scalable SSA techniques has become increasingly urgent. In this paper, we propose a novel multiway extension of SSA, called higher-order multivariate SSA (HO-MSSA), specifically designed for multivariate and multichannel time series signal analysis via tensor decomposition. HO-MSSA utilizes time-delay embedding and tensor singular value decomposition to transform multichannel time series signals into trajectory tensors, which are then decomposed into elementary components in the Fourier domain, rather than the time domain as in traditional SSA methods. These components are grouped into disjoint subsets using spectral clustering, enabling the reconstruction of the underlying source signals. Experimental results demonstrate that HO-MSSA outperforms state-of-the-art SSA methods in various biomedical applications, including electromyography (EMG), electrocardiography (ECG), and electroencephalogram (EEG) signals. • A novel tensor-based multivariate singular spectrum analysis is introduced. • A new time embedding technique embeds time series into a higher dimension. • Tensor SVD is used to factorize the trajectory tensor of time series. • Spectral clustering detects and groups the underlying time series components.

This paper introduces TenSOFO, a novel tensor-based method specifically designed for blind source separation (BSS). Ten-SOFO presents a new efficient alternating direction method of multipliers framework, allowing for simultaneous decomposition of two symmetric third-order tensors under the individual differences in scaling (INDSCAL) format. By establishing a fundamental link between joint INDSCAL decomposition and BSS using second and fourth order statistics, TenSOFO proves to be effective for BSS. The performance of TenSOFO is evaluated in both joint INDSCAL decomposition and BSS tasks, showcasing its remarkable accuracy and potential applications.

In this paper, we investigate the problem of convolutive blind source separation (BSS) via tensor decomposition. A fundamental link between convolutive BSS and block-term decomposition (BTD) is established, forming the basis for our novel tensor-based convolutive BSS method, namely TCBSS. Specifically, the proposed method offers a new effective approach for factorizing tensors under the BTD format where the loading factors are constrained to be identical. By leveraging second-order statistics of data observations, we construct a third-order tensor by stacking covariance matrices at different time lags, and then, apply TCBSS to identify the mixing process. Experimental results demonstrate the robust performance of TCBSS in addressing both BTD and convolutive BSS tasks, particularly when dealing with electromyography (EMG) signal decomposition.

International audience

International audience

International audience

International audience

International audience

International audience

This paper proposes a new event detection algorithm for the use in Non-Intrusive Load Monitoring (NILM). This latter is a field where the main concern is to break down, in a non-intrusive manner, the global electrical energy consumption into individual appliances consumption. Detecting events is thus of importance for appliance clustering in event-based NILM systems. A simple and fast algorithm that detects the variations of the signal's envelope is proposed in this paper. Its main advantage is the high localization accuracy of the start times of events. Its performance is evaluated using simulated and real data and is compared to one of the recently proposed algorithms in the field. Simulations show that the proposed detection algorithm gives 100 % precision and 97.13 % recall at a Signal-to-Noise Ratio (SNR) of 50 dB.

International audience

International audience

We propose to apply the Gaussian Mixture Model (GMM) to surface electromyography (sEMG) signals in order to detect the muscular activation (MA) onset, timing off and intervals. First, classical time and frequency features are extracted from the sEMG signals, beside the Teager-Kaiser energy operator (TKEO) is evaluated and added as a new feature which enhances the detection performance. All the obtained features are then used as the input for the GMM to conduct the binary clustering. Finally, a decision theory is applied in order to declare sEMG activation timing of human skeletal museles during movement. Accuracy and precision of the algorithm are assessed by using a set of synthetic simulated sEMG signals and real ones. A comparison with two previously published techniques is conducted: wavelet transform-based method and double threshold-based method. Our experimental results prove that the proposed GMM-based algorithm is able to accurately reveal the MA timing with performance beyond that of the state-of-the-art methods. Moreover, this proposed algorithm is automatic and user-independent.

The electrical appliances identification problem is gaining a rapidly growing interest these past few years due to the recent need of this information in the new smart grid configuration. In this work, we propose to construct an appliance identification system based on the use of Hidden Markov Models (HMM) to model transient and steady-state electrical current signals. For this purpose, we investigate the usefulness of different choices for the proposed identification system such as: the use of the transient and the steady-state current signals, the use of even and odd-order harmonics as features, and the optimal number of features to take into account. This work also discusses the choice of the Short-Time Fourier Series (STFS) coefficients as adapted features for the representation of transient and steady-state current signals.

In this paper, we investigate the binning problem of joint histogram estimation applied to mutual information based face recognition application. Classical approaches for histograms estimation tend to empirically fix the bin numbers. We evaluate in this work some state of the art rules for automatically choosing the bin numbers. The face recognition problem has been studied in the case of local and holistic methods. The choice’s performance has been evaluated using AT&T database with single sample in the training set. The results show that better accuracy recognition rates can be achieved with data driven bin number choices rather than fixed bin numbers. In the local method, the results show a higher robustness of the automatic vs fixed bin number choice when the regions become smaller.

In this paper, we address the problem of delay detection in mobile localization context. A new methodology for delay detection is introduced, namely the Cell-Averaging Minimum Error Rate CA-MER detector, based on the minimization of the true error probability instead of minimizing only the miss probability for a constant false alarm rate. Simulation results show that the CA-MER detector operates better than the classical ones especially for low SNR values.