Blind Source Separation Techniques Research Articles

Trial-to-Trial Variability in Evoked Neural Responses Exhibits a Very Low Frequency Temporal Signature

In functional magnetic resonance (fMRI) studies, the blood oxygen level dependent (BOLD) signal displays intrinsic spontaneous and task-independent very low frequency (VLF) oscillations (< 0.1 Hz). Most prominent during rest, when they persist into task sessions they can predict trial-to-trial variability in both evoked behavior and brain responses by providing a baseline onto which deterministic responses elicited by the task are superimposed. Moreover, evidence in the literature tentatively suggests that this VLF activity may not be present in the data as distinct, independent source(s) per se, but rather as a mechanism that modulates and perhaps even governs underlying brain processes. Here, we use electrophysiology to investigate the intertrial variability observed in magnetoencephalographic (MEG) event-related field (ERF) components, and to examine whether this variability exhibits a VLF time signature in order to indirectly infer information about the underlying slow waves. The focus is on the visual component, the M100, understood to be regulated by attention. We also explored whether individual differences in the M100 VLF pattern varies as a function of attention deficit/hyperactivity disorder (ADHD) by comparing 11 cases against 11 controls. The M100 component was extracted from the data using a recently introduced blind-source separation technique – space-time independent component analysis (ST-ICA) – which allowed trial-by-trial analysis to be performed on the M100 for proper assessment of VLF modulation. Our results demonstrate, for the first time, the ability of this signal-processing method to isolate relevant components from multidimensional, noisy, ERF data recorded from a highly dense 148-channel MEG system. The intertrial variability in the amplitude and latency of the M100 responses exhibits a slow wave pattern (< 0.1 Hz). However, there was no evidence that the degree of VLF modulation was different in ADHD participants. The role of this VLF activity in brain function is discussed.

Functional source separation improves the quality of single trial visual evoked potentials recorded during concurrent EEG-fMRI

EEG quality is a crucial issue when acquiring combined EEG-fMRI data, particularly when the focus is on using single trial (ST) variability to integrate the data sets. The most common method for improving EEG data quality following removal of gross MRI artefacts is independent component analysis (ICA), a completely blind source separation technique. In the current study, a different approach is proposed based on the functional source separation (FSS) algorithm. FSS is an extension of ICA that incorporates prior knowledge about the signal of interest into the data decomposition. Since in general the part of the EEG signal that will contain the most relevant information is known beforehand (i.e. evoked potential peaks, spectral bands), FSS separates the signal of interest by exploiting this prior knowledge without renouncing the advantages of using only information contained in the original signal waveforms.A reversing checkerboard stimulus was used to generate visual evoked potentials (VEPs) in healthy control subjects. Gradient and ballistocardiogram artefacts were removed with template subtraction techniques to form the raw data, which were then subjected to ICA denoising and FSS. The resulting EEG data sets were compared using several metrics derived from average and ST data and correlated with fMRI data. In all cases, ICA was an improvement on the raw data, but the most obvious improvement was provided by FSS, which consistently outperformed ICA. The results show the benefit of FSS for the recovery of good quality single trial evoked potentials during concurrent EEG-fMRI recordings.

Removing muscle and eye artifacts using blind source separation techniques in ictal EEG source imaging

The contamination of muscle and eye artifacts during an ictal period of the EEG significantly distorts source estimation algorithms. Recent blind source separation (BSS) techniques based on canonical correlation (BSS-CCA) and independent component analysis with spatial constraints (SCICA) have shown much promise in the removal of these artifacts. In this study we want to use BSS-CCA and SCICA as a preprocessing step before the source estimation during the ictal period. Both the contaminated and cleaned ictal EEG were subjected to the RAP-MUSIC algorithm. This is a multiple dipole source estimation technique based on the separation of the EEG in signal and noise subspace. The source estimates were compared with the subtracted ictal SPECT (iSPECT) coregistered to magnetic resonance imaging (SISCOM) by means of the euclidean distance between the iSPECT activations and the dipole location estimates. SISCOM results in an image denoting the ictal onset zone with a propagation. We applied the artifact removal and the source estimation on 8 patients. Qualitatively, we can see that 5 out of 8 patients show an improvement of the dipoles. The dipoles are nearer to or have tighter clusters near the iSPECT activation. From the median of the distance measure, we could appreciate that 5 out of 8 patients show improvement. The results show that BSS-CCA and SCICA can be applied to remove artifacts, but the results should be interpreted with care. The results of the source estimation can be misleading due to excessive noise or modeling errors. Therefore, the accuracy of the source estimation can be increased by preprocessing the ictal EEG segment by BSS-CCA and SCICA. This is a pilot study where EEG source localization in the presurgical evaluation can be made more reliable, if preprocessing techniques such as BSS-CCA and SCICA are used prior to EEG source analysis on ictal episodes.

Removal of the ballistocardiographic artifact from EEG–fMRI data: a canonical correlation approach

The simultaneous recording of electroencephalogram (EEG) and functional magnetic resonance imaging (fMRI) can give new insights into how the brain functions. However, the strong electromagnetic field of the MR scanner generates artifacts that obscure the EEG and diminish its readability. Among them, the ballistocardiographic artifact (BCGa) that appears on the EEG is believed to be related to blood flow in scalp arteries leading to electrode movements. Average artifact subtraction (AAS) techniques, used to remove the BCGa, assume a deterministic nature of the artifact. This assumption may be too strong, considering the blood flow related nature of the phenomenon. In this work we propose a new method, based on canonical correlation analysis (CCA) and blind source separation (BSS) techniques, to reduce the BCGa from simultaneously recorded EEG–fMRI. We optimized the method to reduce the user's interaction to a minimum. When tested on six subjects, recorded in 1.5 T or 3 T, the average artifact extracted with BSS–CCA and AAS did not show significant differences, proving the absence of systematic errors. On the other hand, when compared on the basis of intra-subject variability, we found significant differences and better performance of the proposed method with respect to AAS. We demonstrated that our method deals with the intrinsic subject variability specific to the artifact that may cause averaging techniques to fail.

ICA, kernel methods and nonnegativity: New paradigms for dynamical component analysis of fMRI data

In the last decades, functional magnetic resonance imaging (fMRI) has been introduced into clinical practice. As a consequence of this advanced noninvasive medical imaging technique, the analysis and visualization of medical image time-series data poses a new challenge to both research and medical application. But often, the model data for a regression or generalized linear model-based analysis are not available. Hence exploratory data-driven techniques, i.e. blind source separation (BSS) methods are very popular in functional nuclear magnetic resonance imaging (fMRI) data analysis since they are neither based on explicit signal models nor on a priori knowledge of the underlying physiological process. The independent component analysis (ICA) represents a main BSS method which searches for stochastically independent signals from the multivariate observations. In this paper, we introduce a new kernel-based nonlinear ICA method and compare it to standard BSS techniques. This kernel nonlinear ICA (kICA) overcomes the restrictions of linearity of the mixing process usually encountered with ICA. Dimension reduction is an important preprocessing step for this nonlinear technique and is performed in a novel way: a genetic algorithm is designed which determines the optimal number of basis vectors for a reduced-order feature space representation as an optimization problem of the condition number of the resulting basis. For the fMRI data, a comparative quantitative evaluation is performed between kICA with different kernels, nonnegative matrix factorization (NMF) and other BSS algorithms. The comparative results are evaluated by task-related activation maps, associated time courses and ROC study. The comparison is performed on fMRI data from experiments with 10 subjects. The external stimulus was a visual pattern presentation in a block design. The most important obtained results in this paper represent that kICA and sparse NMF (sNMF) are able to identify signal components with high correlation to the fMRI stimulus, and kICA with a Gaussian kernel is comparable to standard ICA algorithms and even more, it yields spatially focused results.

Separation of single neurons from optical recordings in Tritonia diomedea using ICA

Event Abstract Back to Event Separation of single neurons from optical recordings in Tritonia diomedea using ICA Optical recordings obtained with photodiode arrays and voltage-sensitive dyes allow firing of large numbers of neurons to be recorded during behaviorally-relevant motor programs. Although this technique has tremendous potential for network studies, the resulting datasets can be challenging to interpret; one detector may record multiple neurons and one neuron may appear on many detectors. Earlier studies applied independent component analysis (ICA), a blind source separation technique, to optical traces from Tritonia diomedea's escape swim behavior, to recover components from individual neurons [1] and identify unreported neurons involved in the behavior [2]. Using intracellular and optical methods [3], we evaluate the accuracy of neuronal activity that ICA returns and demonstrate an application of these methods, measuring the change in individual neurons’ activity following stimulus adaptation. To evaluate accuracy, we recorded from neurons in Tritonia with intracellular electrodes while imaging with a 464-element photodiode array using the fast voltage-sensitive absorbance dye RH-155. ICA was run on the optical traces, transforming them into statistically independent components. ICA returned one component for neurons detected by multiple diodes and separated neurons mixed on a single diode. Intracellular recordings confirmed the accuracy of this technique. For each intracellular recording, we found a component that corresponded exactly. Also, the location returned by ICA matched that of the electrode. Additionally, these methods allow us to drive an identified neuron and image its functional connectivity to large numbers of other individual neurons, before and after a treatment of interest. We imaged the responses of several pedal ganglion neurons to a test train of action potentials in CPG neuron C2 driven with an intracellular electrode. Then we applied a sensitizing nerve shock stimulus. Two minutes later we imaged the responses of the same neurons to a second test train. Combining the records and applying ICA, we observed how responses of individual neurons changed after the sensitizing stimulus. Optical recordings in combination with ICA allow quick identification of functional synaptic connections onto previously unknown neurons. Using location maps from ICA, those neurons can be penetrated with intracellular electrodes within the same experiment, enabling new experiments to probe the Tritonia brain over a wide range of conditions and time scales. Given our intracellular validation of ICA's unmixing, these methods promise to be a powerful tool for discovery of unreported neurons and studies of functional connectivity and network plasticity.

Multiple Fault Diagnostics in Solder Paste Printing Process Using Blind Separation Approach

Abstract In this paper, a fault diagnosis and variation pattern identification method is presented based on blind source separation (BSS) approach which is used to extract important information from the measured data and estimate the variation pattern matrix in a solder paste printing process. BSS techniques provide the basis for identifying the precise characteristics of each individual variation pattern in order to facilitate the identification of their root causes. Measurements of mixed signals which collected from a simulated faulty process are decomposed by independent component analysis (ICA) and results from a the simulated case study show that ICA has ability to separate the original source signals from their mixtures and to identify variation pattern characteristics by using frequency domain ICA analysis. The techniques studied in this paper would benefit SMT research of correcting unidentified solder paste stenciling problems.

Intercept Capacity: Unknown Unitary Transformation

We consider the problem of intercepting communications signals between Multiple-Input Multiple-Output (MIMO) communication systems. To correctly detect a transmitted message it is necessary to know the gain matrix that represents the channel between the transmitter and the receiver. However, even if the receiver has knowledge of the message symbol set, it may not be possible to estimate the channel matrix. Blind Source Separation (BSS) techniques, such as Independent Component Analysis (ICA) can go some way to extracting independent signals from individual transmission antennae but these may have been preprocessed in a manner unknown to the receiver. In this paper we consider the situation where a communications interception system has prior knowledge of the message symbol set, the channel matrix between the transmission system and the interception system and is able to resolve the transmissionss from independent antennae. The question then becomes: what is the mutual information available to the interceptor when an unknown unitary transformation matrix is employed by the transmitter.

Improving Mental Task Classification by Adding High Frequency Band Information

Features extracted from delta, theta, alpha, beta and gamma bands spanning low frequency range are commonly used to classify scalp-recorded electroencephalogram (EEG) for designing brain-computer interface (BCI) and higher frequencies are often neglected as noise. In this paper, we implemented an experimental validation to demonstrate that high frequency components could provide helpful information for improving the performance of the mental task based BCI. Electromyography (EMG) and electrooculography (EOG) artifacts were removed by using blind source separation (BSS) techniques. Frequency band powers and asymmetry ratios from the high frequency band (40-100 Hz) together with those from the lower frequency bands were used to represent EEG features. Finally, Fisher discriminant analysis (FDA) combining with Mahalanobis distance were used as the classifier. In this study, four types of classifications were performed using EEG signals recorded from four subjects during five mental tasks. We obtained significantly higher classification accuracy by adding the high frequency band features compared to using the low frequency bands alone, which demonstrated that the information in high frequency components from scalp-recorded EEG is valuable for the mental task based BCI.

Sparse Kernel Independent Component Analysis for Blind Source Separation

We address the problem of Blind Source Separation (BSS) of superimposed signals in situations where one signal has constant or slowly varying intensities at some consecutive locations and at the corresponding locations the other signal has highly varying intensities. Independent Component Analysis (ICA) is a major technique for Blind Source Separation and the existing ICA algorithms fail to estimate the original intensities in the stated situation. We combine the advantages of existing sparse methods and Kernel ICA in our technique, by proposing wavelet packet based sparse decomposition of signals prior to the application of Kernel ICA. Simulations and experimental results illustrate the effectiveness and accuracy of the proposed approach. The approach is general in the way that it can be tailored and applied to a wide range of BSS problems concerning one-dimensional signals and images (two-dimensional signals).

A novel method for multi-fault diagnosis of rotor system

The diagnosis for multiple faults is still a big challenge for monitoring and maintenance of rotating machinery. In this study, the blind source separation (BSS) technique is proposed to separate the vibrational features produced by several faults existing in a rotor. A statistic variable-based algorithm, decorrelation method, is employed to analyze the signals of the rotors with multi-faults, which are simulated by numerical and experimental methods. The separation results show that the typical feature components of the frequency spectrum of each fault are separated successfully. The investigation demonstrates that it is feasible and helpful to adopt BSS technique to diagnose the multi-faults of a rotor.

Adaptive separation of free-surface multiples through independent component analysis

We present a three-stage algorithm for adaptive separation of free-surface multiples. The free-surface multiple elimination (FSME) method requires, as deterministic prerequisites, knowledge of the source wavelet and deghosted data. In their absence, FSME provides an estimate of free-surface multiples that must be subtracted adaptively from the data. First we construct several orders from the free-surface multiple prediction formula. Next we use the full recording duration of any given data trace to construct filters that attempt to match the data and the multiple predictions. This kind of filter produces adequate phase results, but the order-by-order nature of the free-surface algorithm brings results that remain insufficient for straightforward subtraction. Then we construct, trace by trace, a mixing model in which the mixtures are the data trace and its orders of multiple predictions. We separate the mixtures through a blind source separation technique, in particular by employing independent component analysis. One of the recovered signals is a data trace without free-surface multiples. This technique sidesteps the subtraction inherent in most adaptive subtraction methods by separating the desired signal from the free-surface multiples. The method was applied to synthetic and field data. We compared the field data to a published method and found comparable results.

Real-time acoustic blind signal separation system based on the spatio-temporal gradient analysis

This paper presents an autonomous directivity microphone system for the blind source separation based on the newly proposed spatio-temporal gradient algorithm. The blind source separation technique principally uses no a priori knowledge about parameters of convolution, filtering, nor mixing. In the simplest case of the blind source separation problems, the observed mixing signals are the linear combinations of unknown source signals which are mutually and statistically independent and contain zero mean. The proposed blind signal separation algorithm utilizes the linearity among the four signals: (1) the sound pressure, (2) x, (3) y, and (4) z-directional particle velocities, all of which are governed by the wave equation and the equation of motion. The proposed method, therefore, has an ability to simplify the convoluted blind source separation problems into the instantaneous blind source separation problems over the spatio-temporal gradient space. Several acoustical experiments have been performed with acceptable performance of the proposed method of blind source separation. A sensor for measuring the particle velocity of acoustic signal has been used in the experiment.

Compound noise separation in digital circuits using blind source separation

Analysis of individual noise sources in pre-nanometer circuits cannot take into account the evolving reality of multiple noise sources interacting with each other. Noise measurement made at an evaluation node will reflect the cumulative effect of all the active noise sources, while individual and relative severity of various noise sources will determine what types of remedial steps can be taken, pressing the need for development of algorithms that can analyze the contributions of different noise sources when a noise measurement is available. This paper addresses the cocktail-party problem inside integrated circuits with multiple noise sources. It presents a method to extract the time characteristics of individual noise source from the measured compound voltage in order to study the contribution and properties of each source. This extraction is facilitated by application of blind source separation technique, which is based on the assumption of statistical independence of various noise sources over time. The estimated noise sources can aid in performing timing and spectral analysis, and yield better circuit design techniques.

Using blind source separation techniques to improve speech recognition in bilateral cochlear implant patients

Bilateral cochlear implants seek to restore the advantages of binaural hearing by improving access to binaural cues. Bilateral implant users are currently fitted with two processors, one in each ear, operating independent of one another. In this work, a different approach to bilateral processing is explored based on blind source separation (BSS) by utilizing two implants driven by a single processor. Sentences corrupted by interfering speech or speech-shaped noise are presented to bilateral cochlear implant users at 0 dB signal-to-noise ratio in order to evaluate the performance of the proposed BSS method. Subjects are tested in both anechoic and reverberant settings, wherein the target and masker signals are spatially separated. Results indicate substantial improvements in performance in both anechoic and reverberant settings over the subjects' daily strategies for both masker conditions and at various locations of the masker. It is speculated that such improvements are due to the fact that the proposed BSS algorithm capitalizes on the variations of interaural level differences and interaural time delays present in the mixtures of the signals received by the two microphones, and exploits that information to spatially separate the target from the masker signals.

A comparative study of automatic techniques for ocular artifact reduction in spontaneous EEG signals based on clinical target variables: A simulation case

Eye movement artifacts represent a critical issue for quantitative electroencephalography (EEG) analysis and a number of mathematical approaches have been proposed to reduce their contribution in EEG recordings. The aim of this paper was to objectively and quantitatively evaluate the performance of ocular filtering methods with respect to spectral target variables widely used in clinical and functional EEG studies. In particular the following methods were applied: regression analysis and some blind source separation (BSS) techniques based on second-order statistics (PCA, AMUSE and SOBI) and on higher-order statistics (JADE, INFOMAX and FASTICA). Considering blind source decomposition methods, a completely automatic procedure of BSS based on logical rules related to spectral and topographical information was proposed in order to identify the components related to ocular interference. The automatic procedure was applied in different montages of simulated EEG and electrooculography (EOG) recordings: a full montage with 19 EEG and 2 EOG channels, a reduced one with only 6 EEG leads and a third one where EOG channels were not available. Time and frequency results in all of them indicated that AMUSE and SOBI algorithms preserved and recovered more brain activity than the other methods mainly at anterior regions. In the case of full montage: (i) errors were lower than 5% for all spectral variables at anterior sites; and (ii) the highest improvement in the signal-to-artifact (SAR) ratio was obtained up to 40 dB at these anterior sites. Finally, we concluded that second-order BSS-based algorithms (AMUSE and SOBI) provided an effective technique for eye movement removal even when EOG recordings were not available or when data length was short.

Blind Source Separation of Electroencephalographic (EEG) and Electrocorticographic (ECoG) Recordings

Event Abstract Back to Event Blind Source Separation of Electroencephalographic (EEG) and Electrocorticographic (ECoG) Recordings Jason Palmer1*, Arnaud Delorme1, 2, Julie Onton1 and Scott Makeig1 1 University of California San Diego, United States 2 UMR 5549 - CENTRE DE RECHERCHE CERVEAU ET COGNITION, France Global far-field electric field activity of the brain has been studied for some time. The alpha and theta rhythms for example are well known, yet uncertainty remains as to their exact nature and origin. Part of the difficulty in analyzing recordings of the electric activity of the brain is that each electrode records a mixture of a number of sources that are active simultaneously. Apart from brain rhythms, the sensors record electric activity of ocular movements, eye blinks, scalp and facial muscle enervation, heartbeat, and ambient power line signals capacitively coupled to the electrodes. The technique of Independent Component Analysis or Blind Source Separation has proven successful in the separation of these various brain and non-brain sources, providing a significant advance in the ability of neuroscience researchers to zero in on specific brain sources, such as specific sources of brain rhythms, in their analysis, as well as to study the dynamic interactions among the distinguished brain sources. We present a probabilistic framework and an efficient algorithm for performing ICA on large scale EEG, ECoG, or MEG data by minimizing mutual information between the estimated sources. We discuss the relationship between PCA and ICA, and compare a number of ICA algorithms in their ability to reduce mutual information in recorded EEG data, while also producing physiologically plausible brain sources. We find that ICA produces components that are stable over time and consistent across subjects. The ongoing analysis of the sources is expected to greatly advance the current understanding of the nature and origin of the electric activity of the brain.

N‐FindR method versus independent component analysis for lithological identification in hyperspectral imagery

The current study addresses the problem of the identification of each natural material present in each pixel of a hyperspectral image. Two end member extraction methods from hyperspectral imagery were studied: the N‐FindR method and the independent component analysis (ICA). The N‐FindR is an automatic technique that selects extreme points (end members) of an n‐dimensional scatter plot. It assumes the existence of pure pixels in the distribution, which is infrequent in practice. ICA is a blind source separation technique studied in the signal processing community, which allows each spectrum of natural elements (end member spectra) to be extracted from the observation of some linear combinations of these. It considers a more realistic situation than N‐FindR, assuming a spectra mixture for all the pixels. To increase the robustness of ICA, continuum‐removed reflectance spectra were used and an iterative algorithm was introduced that takes into account a major part of the available information. The end member abundances were estimated by the fully constrained least squares spectral mixture analysis (FLCS). The end member identification and quantification were carried out on two surficial formations of a semi arid region located in the Rehoboth region, in Namibia, from hyperspectral Hyperion data. It appears that the two end member extraction methods have a similar potential. Whichever end member extraction method is used, the analysis of the rock abundance maps produces a lot of geological information: the distribution of natural elements is in line with the field observations and allows the description of the formation processes of surficial units.

Blind separation of internal combustion engine vibration signals by a deflation method

Internal combustion engines have several vibration sources, such as combustion, fuel injection, piston slap and valve operation. It is necessary to separate the different vibration sources and then analyze each of them individually. This paper attempts to separate the vibration sources by blind source separation techniques and proposes a combination of the Blind Least Mean Square algorithm with a deflation method to separate several sources. The combined methods are first verified on simulated sources and then applied to engine signals. Separation results show that cylinder pressure consists of two sources—one is a smooth low frequency part due to compression and expansion caused by piston movement; the other is the high frequency part due to combustion.

Piezoelectric Self-Sensing Actuator for Beam Vibration Control Based on Blind Source Separation

A blind source separation based vibration control method using piezoelectric self-sensing actuator is proposed. Two piezoelectric ceramic patches acting as actuators and two current amplifiers consist of the self-sensing bridge circuit. Each piezoelectric current is a mixture of strain rate current and capacitor current. As they are orthogonal and independent with each other, the blind source separation technique is introduced to obtain the pure strain rate signal. An effort is made to solve the problem of BSS uncertainty in separation order. Open loop experimental results on cantilever beam show the feasibility and validity of the proposed method.