Conventional Independent Component Analysis Algorithms Research Articles

High-Throughput Independent Component Analysis Processor for Full Duplex Systems

This paper presents the algorithm and very-large-scale integration (VLSI) architecture of a high-throughput and highly efficient independent component analysis (ICA) processor for self-interference cancellation (SIC) in in-band full-duplex (IBFD) systems. This is the first VLSI architecture reported in the literature based on the state-of-the-art entropy bound minimization (EBM) approach. A novel ICA algorithm is presented in this paper with momentum gradient descent optimization. Simulation results show that the number of iterations for the proposed algorithm is significantly reduced compared to the conventional ICA algorithms. Furthermore, a novel early-distribution estimation scheme is proposed in the designed ICA processor to compute multiple distribution functions with low latency and low complexity. The processing flow and the efficiency for the hardware utilization are specifically designed so that the processing speed is maximized with minimum employment of hardware components. The proposed ICA processor is designed and implemented based on the application-specific-integrated circuit (ASIC) flow. The post-layout estimations show that compared with the conventional EBM-based scheme, the proposed design improves the throughput and efficiency by 30x. In addition, compared to prior designs shown in the literature, the proposed ICA processor also demonstrates a significant enhancement in terms of throughput and efficiency.

Snowball ICA: A Model Order Free Independent Component Analysis Strategy for Functional Magnetic Resonance Imaging Data.

In independent component analysis (ICA), the selection of model order (i.e., number of components to be extracted) has crucial effects on functional magnetic resonance imaging (fMRI) brain network analysis. Model order selection (MOS) algorithms have been used to determine the number of estimated components. However, simulations show that even when the model order equals the number of simulated signal sources, traditional ICA algorithms may misestimate the spatial maps of the signal sources. In principle, increasing model order will consider more potential information in the estimation, and should therefore produce more accurate results. However, this strategy may not work for fMRI because large-scale networks are widely spatially distributed and thus have increased mutual information with noise. As such, conventional ICA algorithms with high model orders may not extract these components at all. This conflict makes the selection of model order a problem. We present a new strategy for model order free ICA, called Snowball ICA, that obviates these issues. The algorithm collects all information for each network from fMRI data without the limitations of network scale. Using simulations and in vivo resting-state fMRI data, our results show that component estimation using Snowball ICA is more accurate than traditional ICA. The Snowball ICA software is available at https://github.com/GHu-DUT/Snowball-ICA.

Single-channel blind signal separation and blind estimation of symbol sequences of communication signals

A new method for the blind separation of multiple communication signals received by a single sensor is presented. In this paper based on the different rate and waveform of communication signals, we employs oversampling and high order statistics to turn a single de-mixing problem into the estimation problem of the waveform and symbol that can be solved by means of a conventional ICA (Independent component analysis) algorithm. Simulation results demonstrate the effectiveness and superiority of the proposed algorithm.

A fast blind source separation algorithm based on the temporal structure of signals

Classical independent component analysis (ICA) has been reasonably successful; however, the performance and the convergence of the conventional ICA algorithms have reached limitations of further improvement since they utilize only the statistical independency among the sources. For circumventing this situation, in this paper, we incorporate some other kinds of temporal priori information, i.e., the generalized autocorrelation and the nonlinear predictability of each source, and make a convex combination of them to formulate a novel cost function for blind source separation (BSS). With this cost function, a fixed-point BSS algorithm is developed. This algorithm inherits the advantages of the well-known FastICA algorithm of ICA, which converges fast and does not need to choose any learning step sizes. Its higher separation accuracy is verified by numerical experiments. Meanwhile, we also give the consistency analysis and prove convergence properties of the algorithm, which has a (locally) consistent estimator and at least quadratic convergence.

Development, validation, and comparison of ICA-based gradient artifact reduction algorithms for simultaneous EEG-spiral in/out and echo-planar fMRI recordings

EEG data acquired in an MRI scanner are heavily contaminated by gradient artifacts that can significantly compromise signal quality. We developed two new methods based on independent component analysis (ICA) for reducing gradient artifacts from spiral in–out and echo-planar pulse sequences at 3 T, and compared our algorithms with four other commonly used methods: average artifact subtraction (Allen, P., Josephs, O., Turner, R., 2000. A method for removing imaging artifact from continuous EEG recorded during functional MRI. NeuroImage 12, 230–239.), principal component analysis (Niazy, R., Beckmann, C., Iannetti, G., Brady, J., Smith, S., 2005. Removal of FMRI environment artifacts from EEG data using optimal basis sets. NeuroImage 28, 720–737.), Taylor series ( Wan, X., Iwata, K., Riera, J., Kitamura, M., Kawashima, R., 2006. Artifact reduction for simultaneous EEG/fMRI recording: adaptive FIR reduction of imaging artifacts. Clin. Neurophysiol. 117, 681–692.) and a conventional temporal ICA algorithm. Models of gradient artifacts were derived from simulations as well as a water phantom and performance of each method was evaluated on datasets constructed using visual event-related potentials (ERPs) as well as resting EEG. Our new methods recovered ERPs and resting EEG below the beta band (<12.5 Hz) with high signal-to-noise ratio (SNR>4). Our algorithms outperformed all of these methods on resting EEG in the theta and alpha bands (SNR>4); however, for all methods, signal recovery was modest (SNR∼1) in the beta band and poor (SNR<0.3) in the gamma band and above. We found that the conventional ICA algorithm performed poorly with uniformly low SNR (<0.1). Taken together, our new ICA-based methods offer a more robust technique for gradient artifact reduction when scanning at 3 T using spiral in–out and echo-planar pulse sequences. We provide new insights into the strengths and weaknesses of each method using a unified subspace framework.

Time-Domain Blind Signal Separation of Convolutive Mixtures via Multidimensional Independent Component Analysis

An algorithm for blind signal separation (BSS) of convolutive mixtures is presented. In this algorithm, the BSS problem is treated as multidimensional independent component analysis (ICA) by introducing an extended signal vector which is composed of current and previous samples of signals. It is empirically known that a number of conventional ICA algorithms solve the multidimensional ICA problem up to permutation and scaling of signals. In this paper, we give theoretical justification for using any conventional ICA algorithm. Then, we discuss the remaining problems, i.e., permutation and scaling of signals. To solve the permutation problem, we propose a simple algorithm which classifies the signals obtained by a conventional ICA algorithm into mutually independent subsets by utilizing temporal structure of the signals. For the scaling problem, we prove that the method proposed by Koldovský and Tichavský is theoretically proper in respect of estimating filtered versions of source signals which are observed at sensors.

Novel statistical approach to blind recovery of earth signal and source wavelet using independent component analysis

This paper provides a new statistical approach to blind recovery of both earth signal and source wavelet given only the seismic traces using independent component analysis (ICA) by explicitly exploiting the sparsity of both the reflectivity sequence and the mixing matrix. Our proposed blind seismic deconvolution algorithm consists of three steps. Firstly, a transformation method that maps the seismic trace convolution model into multiple inputs multiple output (MIMO) instantaneous ICA model using zero padding matrices has been proposed. As a result the nonzero elements of the sparse mixing matrix contain the source wavelet. Secondly, whitening the observed seismic trace by incorporating the zero padding matrixes is conducted as a pre-processing step to exploit the sparsity of the mixing matrix. Finally, a novel logistic function that matches the sparsity of reflectivity sequence distribution has been proposed and fitted into the information maximization algorithm to obtain the demixing matrix. Experimental simulations have been accomplished to verify the proposed algorithm performance over conventional ICA algorithms such as Fast ICA and JADE algorithm. The mean square error (MSE) of estimated wavelet and estimated reflectivity sequence shows the improvement of proposed algorithm.

Geometrically Constrained Independent Component Analysis

Acoustical signals are often corrupted by other speeches, sources, and background noise. This makes it necessary to use some form of preprocessing so that signal processing systems such as a speech recognizer or machine diagnosis can be effectively employed. In this contribution, we introduce and evaluate a new algorithm that uses independent component analysis (ICA) with a geometrical constraint [constrained ICA (CICA)]. It is based on the fundamental similarity between an adaptive beamformer and blind source separation with ICA, and does not suffer the permutation problem of ICA-algorithms. Unlike conventional ICA algorithms, CICA needs prior knowledge about the rough direction of the target signal. However, it is more robust against an erroneous estimation of the target direction than adaptive beamformers: CICA converges to the right solution as long as its look direction is closer to the target signal than to the jammer signal. A high degree of robustness is very important since the geometrical prior of an adaptive beamformer is always roughly estimated in a reverberant environment, even when the look direction is precise. The effectiveness and robustness of the new algorithms is proven theoretically, and shown experimentally for three sources and three microphones with several sets of real-world data

分類情報に基づく教師あり独立成分分析による特徴抽出

Independent component analysis (ICA) is a technique of transforming observation signals into their unknown independent components; hence, ICA has often been applied to blind signal separation problems. In this application, it is expected that the obtained independent components extract essential information of independent signal sources from input data in an unsupervised fashion. Based on such characteristics, ICA is currently utilized as a feature extraction method for images and sounds for recognition purposes. However, since ICA is an unsupervised learning, the obtained independent components are not always useful in recognition. To overcome this problem, we propose a supervised approach to ICA using category information. The proposed method is implemented in a conventional three-layered neural network, but its objective function to be minimized is defined not only for the output layer but also for the hidden layer. The objective function consists of the following two terms: one evaluates the kurtosis of hidden unit outputs and the other evaluates the error between output signals and their teacher signals. The experiments are performed using several standard datasets to evaluate performance of the proposed algorithm. It is confirmed that a higher recognition accuracy is attained by the proposed method as compared with a conventional ICA algorithm. © 2007 Wiley Periodicals, Inc. Electr Eng Jpn, 161(2): 25–32, 2007; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/eej.20522

Source density-driven independent component analysis approach for fMRI data.

Independent component analysis (ICA) has become a popular tool for functional magnetic resonance imaging (fMRI) data analysis. Conventional ICA algorithms including Infomax and FAST-ICA algorithms employ the underlying assumption that data can be decomposed into statistically independent sources and implicitly model the probability density functions of the underlying sources as highly kurtotic or symmetric. When source data violate these assumptions (e.g., are asymmetric), however, conventional ICA methods might not work well. As a result, modeling of the underlying sources becomes an important issue for ICA applications. We propose a source density-driven ICA (SD-ICA) method. The SD-ICA algorithm involves a two-step procedure. It uses a conventional ICA algorithm to obtain initial independent source estimates for the first-step and then, using a kernel estimator technique, the source density is calculated. A refitted nonlinear function is used for each source at the second step. We show that the proposed SD-ICA algorithm provides flexible source adaptivity and improves ICA performance. On SD-ICA application to fMRI signals, the physiologic meaningful components (e.g., activated regions) of fMRI signals are governed typically by a small percentage of the whole-brain map on a task-related activation. Extra prior information (using a skewed-weighted distribution transformation) is thus additionally applied to the algorithm for the regions of interest of data (e.g., visual activated regions) to emphasize the importance of the tail part of the distribution. Our experimental results show that the source density-driven ICA method can improve performance further by incorporating some a priori information into ICA analysis of fMRI signals.

クラス毎の独立成分を用いたパターン認識方法

Applications of independent component analysis (ICA) to feature extraction have been a topic of research interest. However, the effectiveness of pattern features extracted by conventional ICA algorithms greatly depends on datasets in general. As one of the reasons, we have pointed out that conventional ICA features are obtained by increasing only their independence even if class information is available. In this paper, we propose a supervised learning approach to ICA to extract useful and robust features. The proposed method consists of several modules, each of which is responsible for extracting features for each class and identifying the class labels using the k nearest neighbor classifier. All the module outputs are combined to identify final results based on a majority rule. We evaluate the performance of the proposed method in several recognition tasks. From these results, we confirm the effectiveness of the recognition method using independent components for each class.

クラス間距離を最大化する教師あり独立成分分析による特徴抽出

Recently, Independent Component Analysis (ICA) has been applied to not only problems of blind signal separation, but also feature extraction of patterns. However, the effectiveness of features extracted by ICA (ICA features) has not been verified yet. As one of the reasons, it is considered that ICA features are obtained by increasing their independence rather than by increasing their class separability. Hence, we can expect that high-performance pattern features are obtained by introducing supervisor into conventional ICA algorithms such that the class separability of features is enhanced.. In this work, we propose SICA by maximizing Mahalanobis distance between classes. Moreover, we propose a new distance measure in which each ICA feature is weighted by the power of principal components consisting of the ICA feature. In the recognition experiments, we demonstrate that the better recognition accuracy for two data sets in UCI Machine Learning Repository is attained when using features extracted by the proposed SICA.

Differential Hebbian-type learning algorithms for decorrelation and independent component analysis

Differential learning algorithms for decorrelation and independent component analysis (ICA) are presented. It is shown that the proposed differential Hebbian-type learning algorithms are able to successfully decorrelate the non-zero mean-valued data without any preprocessing. Differential learning is also applied for independent component analysis (ICA) so that non-zero mean-valued source signals can be recovered without any preprocessing. It is demonstrated that modified ICA algorithms using differential learning have a superior performance compared to conventional ICA algorithms for the case where the mean values of source signals are non-zero and are changing.