Search

This paper presents a new dataset of measured multichannel room impulse responses (RIRs) named dEchorate. It includes annotations of early echo timings and 3D positions of microphones, real sources, and image sources under different wall configurations in a cuboid room. These data provide a tool for benchmarking recent methods in echo-aware speech enhancement, room geometry estimation, RIR estimation, acoustic echo retrieval, microphone calibration, echo labeling, and reflector position estimation. The dataset is provided with software utilities to easily access, manipulate, and visualize the data as well as baseline methods for echo-related tasks.

The purpose of this study was to enhance the performance of steady-state visual evoked potential (SSVEP)-based visual acuity assessment with spatial filtering methods. Using the vertical sinusoidal gratings at six spatial frequency steps as the visual stimuli for 11 subjects, SSVEPs were recorded from six occipital electrodes (O1, Oz, O2, PO3, POz, and PO4). Ten commonly used training-free spatial filtering methods, i.e., native combination (single-electrode), bipolar combination, Laplacian combination, average combination, common average reference (CAR), minimum energy combination (MEC), maximum contrast combination (MCC), canonical correlation analysis (CCA), multivariate synchronization index (MSI), and partial least squares (PLS), were compared for multielectrode signals combination in SSVEP visual acuity assessment by statistical analyses, e.g., Bland–Altman analysis and repeated-measures ANOVA. The SSVEP signal characteristics corresponding to each spatial filtering method were compared, determining the chosen spatial filtering methods of CCA and MSI with a higher performance than the native combination for further signal processing. After the visual acuity threshold estimation criterion, the agreement between the subjective Freiburg Visual Acuity and Contrast Test (FrACT) and SSVEP visual acuity for the native combination (0.253 logMAR), CCA (0.202 logMAR), and MSI (0.208 logMAR) was all good, and the difference between FrACT and SSVEP visual acuity was also all acceptable for the native combination (−0.095 logMAR), CCA (0.039 logMAR), and MSI (−0.080 logMAR), where CCA-based SSVEP visual acuity had the best performance and the native combination had the worst. The study proved that the performance of SSVEP-based visual acuity can be enhanced by spatial filtering methods of CCA and MSI and also recommended CCA as the spatial filtering method for multielectrode signals combination in SSVEP visual acuity assessment.

Vibration-based local damage detection in rotating machines (i.e., rolling element bearings) is typically a problem of detecting low-energy cyclic impulsive modulations in the measured signal. This can be challenging as both the amplitude of a single damage-related impulse and the distance between impulses might be changing in time. From the signal processing point of view, this means time varying regarding the the signal-to-noise ratio, location of information in the frequency domain, and loss of periodicity (this remains cyclic but not periodic). One of the many attempted approaches to this problem is filtration using custom filters derived in a data-driven fashion. One of the methods to obtain such filters is a selector approach, where the value of a certain statistic is calculated for individual frequency bands of a signal that results in the magnitude response of a filter. In this approach, each chosen statistic will yield different results, and the obtained filter will be focused on different frequency bands focusing on different behaviors. One of the most popular and powerful selectors is spectral kurtosis as popularized by Antoni, which uses kurtosis as an operational statistic. Unfortunately, after closer inspection, it is easy to notice that, although selectors can significantly enhance the signal, they accumulate a great deal of noise and other background content of signals, which occupies the space (or rather time) in between the impulses. Another disadvantage is that such filters are time-invariant, because, in the principle of their construction, they are not adaptive, and even slight changes in the signal yield suboptimal results either by missing relevant data when the conditions in the signal change (i.e., informative impulses widen in bandwidth), or by accumulating unnecessary noise when the relevant information is not present (in between impulses or in frequency bands that impulses no longer occupy). To address this issue, I propose generalization of the selector approach using the example of spectral kurtosis. This assumes creating a time-varying selector that can be seen as a spatial filter in the time-frequency domain. The time-varying SK (TVSK) is estimated for segments of the signal, and, instead of a vector of SK-based filter coefficients, one obtains a TVSK-based matrix of coefficients that takes into account the time-varying properties of the signal. The obtained structure is then binarized and used as a filter. The presented method is tested using a simulated signal as well as two real-life signals measured on heavy-duty bearings in two different types of machine.

In this paper, an efficient sidelobe levels (SLL) reduction and spatial filtering algorithm is proposed for linear one-dimensional arrays. In this algorithm, the sidelobes are beamspace processed simultaneously based on its orientation symmetry to achieve very deep SLL at much lower processing time compared with recent techniques and is denoted by the sidelobes simultaneous reduction (SSR) algorithm. The beamwidth increase due to SLL reduction is found to be the same as that resulting from the Dolph-Chebyshev window but at considerably lower average SLL at the same interelement spacing distance. The convergence of the proposed SSR algorithm can be controlled to guarantee the achievement of the required SLL with almost steady state behavior. On the other hand, the proposed SSR algorithm has been examined for spatial selective sidelobe filtering and has shown the capability to effectively reduce any angular range of the radiation pattern effectively. In addition, the controlled convergence capability of the proposed SSR algorithm allows it to work at any interelement spacing distance, which ranges from tenths to a few wavelength distances, and still provide very low SLL.

This thesis explores machine learning models for the analysis and classification of electroencephalographic (EEG) signals used in Brain-Computer Interface (BCI) systems. The goal is 1) to develop a system that allows users to control home-automation devices using their mind, and 2) to investigate whether it is possible to achieve this, using low-cost EEG equipment. The thesis includes both a theoretical and a practical part. In the theoretical part, we overview the underlying principles of Brain-Computer Interface systems, as well as, different approaches for the interpretation and the classification of brain signals. We also discuss the emergent launch of low-cost EEG equipment on the market and its use beyond clinical research. We then dive into more technical details that involve signal processing and classification of EEG patterns using machine leaning. Purpose of the practical part is to create a brain-computer interface that will be able to control a smart home environment. As a first step, we investigate the generalizability of different classification methods, conducting a preliminary study on two public datasets of brain encephalographic data. The obtained accuracy level of classification on 9 different subjects was similar and, in some cases, superior to the reported state of the art. Having achieved relatively good offline classification results during our study, we move on to the last part, designing and implementing an online BCI system using Python. Our system consists of three modules. The first module communicates with the MUSE (a low-cost EEG device) to acquire the EEG signals in real time, the second module process those signals using machine learning techniques and trains a learning model. The model is used by the third module, that takes control of cloud-based home automation devices. Experiments using the MUSE resulted in significantly lower classification results and revealed the limitations of the low-cost EEG signal acquisition device for online BCIs. &nbsp

Radio astronomical observations are increasingly contaminated by RF interference. Assuming an array of telescopes, a previous technique considered spatial filtering based on projecting out the interferer array signature vector. A disadvantage is that this effectively reduces the array by one (expensive) telescope. In this paper, we consider extending the astronomical array with a reference antenna array, and develop spatial filtering algorithms for this situation. The information from the reference antennas improves the quality of the interferer signature vector estimation, hence more of the interference can be projected out. Moreover, since only the covariance data of the astronomical array has to be reconstructed, the conditioning of the problem improves as well. The algorithms are tested both on simulated and experimental data.

International audience; Objective: Most current Electroencephalography (EEG)-based Brain-Computer Interfaces (BCIs) are based on machine learning algorithms. There is a large diversity of classifier types that are used in this field, as described in our 2007 review paper. Now, approximately 10 years after this review publication, many new algorithms have been developed and tested to classify EEG signals in BCIs. The time is therefore ripe for an updated review of EEG classification algorithms for BCIs. Approach: We surveyed the BCI and machine learning literature from 2007 to 2017 to identify the new classification approaches that have been investigated to design BCIs. We synthesize these studies in order to present such algorithms, to report how they were used for BCIs, what were the outcomes, and to identify their pros and cons. Main results: We found that the recently designed classification algorithms for EEG-based BCIs can be divided into four main categories: adaptive classifiers, matrix and tensor classifiers, transfer learning and deep learning, plus a few other miscellaneous classifiers. Among these, adaptive classifiers were demonstrated to be generally superior to static ones, even with unsupervised adaptation. Transfer learning can also prove useful although the benefits of transfer learning remain unpredictable. Riemannian geometry-based methods have reached state-of-the-art performances on multiple BCI problems and deserve to be explored more thoroughly, along with tensor-based methods. Shrinkage linear discriminant analysis and random forests also appear particularly useful for small training samples settings. On the other hand, deep learning methods have not yet shown convincing improvement over state-of-the-art BCI methods. Significance: This paper provides a comprehensive overview of the modern classification algorithms used in EEG-based BCIs, presents the principles of these Review of Classification Algorithms for EEG-based BCI 2 methods and guidelines on when and how to use them. It also identifies a number of challenges to further advance EEG classification in BCI.

A multipath mitigation technique for GNSS is proposed.The technique exploites the change of polarization when a signal is reflected.A novel statistic based on the outputs of a correlator bank is introduced to estimate the multipath DoA and polarization.Applying the technique increases the estimation performance in the case of highly temporally or spatially correlated multipath and line-of-sight signal. This paper treats the problem of line-of-sight (LOS) parameter estimation in strong multipath environments. In the case of highly temporally and spatially correlated LOS and multipath signals, such as urban canyons, common multipath mitigation methods are highly degraded, as signal separation cannot be performed in the spatio temporal domain. In this case, we exploit the LOS and multipath polarization diversity to decouple the signals using an antenna array with right-hand-circular polarization (RHCP) and left-hand-circular polarization (LHCP) feeds. The multipath direction- of-arrival (DOA) and polarization coefficients can effectively be estimated from the LHCP spatial covariance matrix. The LOS DOA can be estimated from the RHCP spatial covariance matrix. The spatial covariance matrices are calculated from the outputs of a matched correlator bank. The DOA and polarization estimates are used to implement a dual polarization beamformer, which maximizes the LOS energy and suppresses multipath energy over both polarizations. The LOS time-delay is estimated from the beamformer output with a maximum-likelihood estimator with a significantly reduced number of parameters and computational complexity in comparison to a full model estimator. Simulation results for GPS show that the proposed dual polarization beamforming algorithm performs better than an equivalent single-polarization beamformer in a dense multipath environment.

International audience; In this paper, the estimation of spatio-temporal patterns in the context of event-related potentials or evoked potentials studies in neuroscience is addressed. The proposed framework (denoted xDAWN) has the advantage to require only the knowledge of the time of stimuli onsets which are determined by the experimental setup. A theoretical analysis of the xDAWN framework shows that it provides asymptotically optimal spatial filters under weak assumptions. The loss in signal to interference-plus-noise ratio due to finite sample effect is calculated in a closed form at the first order of perturbation and is then validated by simulations. This last result shows that the proposed method provides interesting performance and outperforms classical methods, such as independent component analysis, in a wide range of situations. Moreover, the xDAWN algorithm has the property to be robust with respect to the model parameter values. Finally, validations on real electro-encephalographic data confirm the good behavior of the proposed xDAWN framework in the context of a P300 speller brain-computer interface.

Hearing aids are electronic, battery-operated sensing devices which aim at compensating various kinds of hearing impairments. Recent advances in low-power electronics coupled with progresses made in digital signal processing offer the potential for substantial improvements over state-of-the-art systems. Nevertheless, efficient noise reduction in complex listening scenarios remains a challenging task, partly due to the limited number of microphones that can be integrated on such devices. We investigate the noise reduction capability of hearing instruments that may exchange data by means of a rate-constrained wireless link and thus benefit from the signals recorded at both ears of the user. We provide the necessary theoretical results to analyze this collaboration mechanism under two different coding strategies. The first approach takes full benefit of the binaural correlation, while the second neglects it, since binaural statistics are difficult to estimate in a practical setting. The gain achieved by collaborating hearing aids as a function of the communication bit rate is then characterized, both in a monaural and a binaural configuration. The corresponding optimal rate allocation strategies are computed in closed form. While the analytical derivation is limited to a simple acoustic scenario, the latter is shown to capture many of the features of the general problem. In particular, it is observed that the loss incurred by coding schemes which do not consider the binaural correlation is rather negligible in a very noisy environment. Finally, numerical results obtained using real measurements corroborate the potential of our approach in a realistic scenario.

This paper proposes and analyzes two different linear space-frequency architectures for the reception of OFDM-modulated signals: the classical sample matrix inversion (SMI) algorithm and a new architecture that maximizes the output signal-to-interference-plus noise ratio (MSINR). The performance of these two linear receivers is compared in terms of asymptotic output SINR, taking into account the finite sample size effect through the asymptotic covariance of the filter weights. The analysis is asymptotic in the sense that the performance is analyzed, assuming that both the number of carriers and the prefix length of the OFDM signal increase without bound at the same rate, whereas their quotient remains constant. Assuming that the carrier frequencies become asymptotically close to one another, we are able to derive explicit equations that shed some light on the influence of the frequency selectivity of channel and interference on the relative performance of the two approaches. The results are useful in the sense that they provide first-order approximations to the (asymptotically) optimum number of adjacent carriers to be processed by a single beamformer in a finite sample size situation.

This paper proposes a spatial filtering technique for the reception of pilot-aided multirate multicode direct-sequence code division multiple access (DS/CDMA) systems such as wideband CDMA (WCDMA). These systems introduce a code-multiplexed pilot sequence that can be used for the estimation of the filter weights, but the presence of the traffic signal (transmitted at the same time as the pilot sequence) corrupts that estimation and degrades the performance of the,filter significantly. This is caused by the fact that although the traffic and pilot signals are usually designed to be orthogonal, the frequency selectivity of the channel degrades this orthogonality at hte receiving end. Here, we propose a semi-blind technique that eliminates the self-noise caused by the code-multiplexing of the pilot. We derive analytically the asymptotic performance of both the training-only and the semi-blind techniques and compare them with the actual simulated performance. It is shown, both analytically and A. a simulation, that high gains can be achieved with respect to training-only-based techniques.

Despite the well known advantages that a space-variant representation of the visual signal offers, the required adaptation of the algorithms developed in the Cartesian domain, before applying them in the log-polar space, has limited a wide use of such representation in visual processing applications. In this paper, we present a set of original rules for designing a discrete log-polar mapping that allows a direct application in the log-polar domain of the algorithms, based on spatial multi-scale and multi-orientation filtering, originally developed for the Cartesian domain. The advantage of the approach is to gain, without modifications, an effective space-variance and data reduction. Such design strategies are based on a quantitative analysis of the relationships between the spatial filtering and the space-variant representation. We assess the devised rules by using a distributed approach based on a bank of band-pass filters to compute reliable disparity maps, by providing quantitative measures of the computational load and of the accuracy of the computed visual features.

International audience; The Computis European project (2006-2009) was aimed at developing innovative experimental mass spectrometry imaging (MSI) techniques and software tools for data treatment and visualization, and at validating them in key biological applications (neurobiology, pharmaceutical drug development). The project succeeded in defining a specific standard format for MSI, imzML, in collaboration with the HUPO consortium. The four main MSI software tools developed by the project all handle the imzML format.Data Cube Explorer provides an easy spectral and spatial exploration of MS images: spectrum zooming, scrolling through the dataset masses with a manual contrast tuning for images, selection of Regions Of Interest with the display of the associated spectra. The self-organizing map functionality classifies images according to the intensity of all pixel places and automatically selects images as different as possible. Mirion is a simple visualization module displaying spectra by pixel and for the total image, with zooming and scrolling functions. Histogram of the total ion count of each pixel can be calculated, using different input parameters. Images are displayed for each peak of the total spectrum, with a manual intensity tuning and a comparison of the intensity distributions by pixel between several images.EasyMSI enables spatial and spectral visualization of mass spectrometry imaging datasets (spectrum and image display, peak and pixel picking, zooming on spectra and images, ROI selection), as well as an assistance for the interpretation of data: This assistance includes indicators (relative variance, Moran index, m/z correlation) to highlight peaks that bring interesting information, peak list for molecule identification, spectrum denoising or structure analysis by clustering methods (K-means, fuzzy, hierarchical clustering, diffusion map). EasyMSI offers the advantage of processing and displaying the original data (i.e. without binning).BioMap is an image analysis platform for MSI and Magnetic Resonance Imaging. It includes viewing functions (spectrum and image display, intensity adjustment, zoom, treatment of multiple ROIs, geometrical transformations and operations), and spectrum treatment (spatial or temporal filtering, baseline correction, detrending). More elaborated functions enable a simultaneous view of all dataset images, creation of a movie, statistical and histogram analysis, co-registration of images of one or two dataset(s), and realignment of images. The use and capacities of these tools are presented through a comparative analysis of a rodent urinary bladder dataset in imzML format.

Non-destructive testing (NDT) techniques are critical to ensure integrity and safety of engineered structures. Structural health monitoring (SHM) is considered as the next step in the field enabling continuous monitoring of structures. The first part of the thesis concerns NDT and SHM using guided waves in plates, or Lamb waves, to perform imaging of plate structures. The imaging is performed using a fixed active array setup covering a larger area of a plate. Current methods are based on conventional beamforming techniques that do not efficiently exploit the available data from the small arrays used for the purpose. In this thesis an adaptive signal processing approach based on the minimum variance distortionless response (MVDR) method is proposed to mitigate issues related to guided waves, such as dispersion and the presence of multiple propagating modes. Other benefits of the method include a significant increase in resolution. Simulation and experimental results show that the method outperforms current standard processing techniques. The second part of the thesis addresses transducer design issues for resonant ultrasound inspections. Resonant ultrasound methods utilize the shape and frequency of the object's natural modes of vibration to detect anomalies. The method considered in the thesis uses transducers that are acoustically coupled to the inspected structures. Changes in the transducer's electrical impedance are used to detect defects. The sensitivity that can be expected from such a setup is shown to highly depend on the transducer resonance frequency, as well as the working frequency of the instrument. Through simulations and a theoretical argumentation, optimal conditions to achieve high sensitivity are given.

This paper proposes a spatial filtering technique for the reception of pilot-aided multirate multicode direct-sequence code division multiple access (DS/CDMA) systems such as wideband CDMA (WCDMA). These systems introduce a code-multiplexed pilot sequence that can be used for the estimation of the filter weights, but the presence of the traffic signal (transmitted at the same time as the pilot sequence) corrupts that estimation and degrades the performance of the filter significantly. This is caused by the fact that although the traffic and pilot signals are usually designed to be orthogonal, the frequency selectivity of the channel degrades this orthogonality at hte receiving end. Here, we propose a semi-blind technique that eliminates the self-noise caused by the code-multiplexing of the pilot. We derive analytically the asymptotic performance of both the training-only and the semi-blind techniques and compare them with the actual simulated performance. It is shown, both analytically and via simulation, that high gains can be achieved with respect to training-onlybased techniques.