Rafael Ferrari

Unsupervised Processing of Geophysical Signals: A Review of Some Key Aspects of Blind Deconvolution and Blind Source Separation

This article reviews some key aspects of two important branches in unsupervised signal processing: blind deconvolution and blind source separation (BSS). It also gives an overview of their potential application in seismic processing, with an emphasis on seismic deconvolution. Finally, it presents illustrative results of the application, on both synthetic and real data, of a method for seismic deconvolution that combines techniques of blind deconvolution and blind source separation. Our implementation of this method contains some improvements overthe original method in the literature described.

Unsupervised channel equalization using fuzzy prediction-error filters

Ee present a new paradigm for unsupervised nonlinear equalization based on prediction-error fuzzy filters. Tests in different linear channel scenarios are carried out in order to assess the performance of the equalizer. The results show that the proposal is solid and may provide a performance close to that of a Bayesian equalizer.

Nonlinear Blind Source Deconvolution Using Recurrent Prediction-Error Filters and an Artificial Immune System

In this work, we propose a general framework for nonlinear prediction-based blind source deconvolution that employs recurrent structures (multi-layer perceptrons and an echo state network) and an immune-inspired optimization tool. The paradigm is tested under different channel models and, in all cases, the presence of feedback loops is shown to be a relevant factor in terms of performance. These results open interesting perspectives for dealing with classical problems such as equalization and blind source separation.

Achievable rates of DSL with crosstalk cancellation

Crosstalk is one of the main limiting factors in the data rates achievable by digital subscriber line (DSL) systems, and several algorithms have been proposed to mitigate this impairment. In this paper, we compare the achievable rates of binders under different crosstalk-mitigating techniques. When computing these rates, we also compare two different power constraints: either on the total power in the binder or on the power in each twisted wire pair. We will see that, for the scenarios considered in this paper, the fact that the signals are jointly processed in one or both ends of the DSL link leads to roughly the same performance, which can be far superior to that of systems with no cooperation between the users. Both power constraints also lead to similar achievable rates. Copyright

A nonlinear prediction approach to the blind separation of convolutive mixtures

We propose a method for source separation of convolutive mixture based on nonlinear prediction-error filters. This approach converts the original problem into an instantaneous mixture problem, which can be solved by any of the several existing methods in the literature. We employ fuzzy filters to implement the prediction-error filter, and the efficacy of the proposed method is illustrated by some examples.

MLP-Based Equalization and Pre-Distortion Using an Artificial Immune Network

Due to its universal approximation capability, the multilayer perceptron (MLP) neural network has been applied to several function approximation and classification tasks. Despite its success in solving these problems, its training, when performed by a gradient-based method, is sometimes hindered by the existence of unsatisfactory solutions (local minima). In order to overcome this difficulty, this paper proposes a novel approach to the training of a MLP based on a simple artificial immune network model. The application domain for assessing the performance of the proposed technique is that of digital communications, in particular, the problems of channel equalization and pre-distortion. The obtained simulation results demonstrate that the proposal is capable of efficiently solving the problems tackled

The capacity of binders for MIMO digital subscriber lines

Crosstalk is one of the main limiting factors in the data rates achievable by digital subscriber line (DSL) systems, and several algorithms have been proposed to mitigate this impairment. In this paper, we compare the capacity of binders under different crosstalk-mitigating techniques. When computing capacity, we also compare two different power constraints: either on the total power in the binder or on the power in each pair. We will see that, for the scenarios considered in this paper, the fact that the signals are jointly processed in one or both ends of the DSL link leads to roughly the same performance, which is far superior to that of systems with no cooperation between the users. Both power constraints also lead to similar achievable rates.

A comparative study of non-MSE criteria in nonlinear equalization

This work studies the application of non-MSE criteria to adapt the linear readout of Extreme Learning Machines (ELMs) in the context of communication channel equalization. A qualitative and experimental analysis is performed, in terms of bit error rate, optimization surface and decision boundary. The results reached by the ELM-based equalizer, considering three different noise models, did not reveal clear advantages of using criteria based on the concepts of error entropy, correntropy, and the L1-norm of the error. Notwithstanding, the observed results motivate a theoretical investigation on the conditions under which the potential discrepancies between the optimal solutions of these criteria may be stressed.

Support Vector Clustering Applied to Digital Communications

Support vector clustering (SVC) is a recently proposed clustering methodology with promising performance for high-dimensional and noisy datasets, and for clusters with arbitrary shape. This work addresses the application of SVC, a kernel-based method, in a context in which the channel equalization problem is conceived as a clustering task. The main challenge, in this case, is to perform unsupervised clustering aiming at the design of an optimal Bayesian or a blind prediction-based receiver without resorting to a priori information about the transmission medium. The proposed technique employs a two-stage procedure -a combination between the use of SVC to obtain a first set of clusters and an auxiliary heuristic to help separating eventual multiple clouds contained in a single cluster and attribute centers to them via an iterated local search (ILS) algorithm. The obtained results indicate that kernel methods can be successfully applied to the field of signal processing.

A Hybrid Unsupervised Clustering Algorithm for Channel Equalization

In this work, we propose and analyze the applicability of a novel unsupervised data clustering technique in the problem of channel equalization. The proposal combines two different methods, a neuro-immune network called RABNET [1] and the iterated local search algorithm (ILS) [2], to produce a tool that, in contrast to classical solutions like the k-means algorithm, does not require a priori knowledge about the number of clusters to be found and, moreover, possesses mechanisms to avoid local convergence. Simulation results attest both the viability and efficiency of the proposal in scenarios conceived to highlight certain aspects that can be decisive insofar as real-world applications are concerned.