Paulo B. Batalheiro

Nonuniform Subband Adaptive Filtering With Critical Sampling

Adaptive subband structures have been proposed with the objective of increasing the convergence speed and/or reducing the computational complexity of conventional adaptive algorithms, mainly for applications that require a large number of adaptive coefficients. In this paper, we present a nonuniform subband structure with critical sampling, which is capable of modeling an arbitrary finite-impulse response (FIR) system with reduced aliasing. A least-mean-square (LMS)-type adaptation algorithm with normalized step sizes, which works at the lowest downsampling rate and minimizes the average of the subband squared errors, is derived for the proposed structure. A convergence analysis of the adaptation algorithm is presented, from which its convergence rate and steady-state mean-square error can be estimated.

Improving the near-perfect hybrid filter bank performance in the presence of realization errors

Hybrid filter banks have received increasing attention in the literature, for applications such as high-speed, high-resolution A/D and D/A converter design. In the manufacturing process, however, the, filter coefficients of a hybrid filter bank are plagued with some errors due to technological limitations, particularly those of the analog filters, leading to degradation of the system performance. This work presents a novel method for improving the mean signal-to-noise-ratio of near-perfect reconstruction filter banks, taking into account such realization errors. The method consists in minimizing the total noise energy derived in an accurate way by a theoretical expression.

Non-uniform subband adaptive filtering with critical sampling

Adaptive subband structures have been proposed with the objective of increasing the convergence speed and/or reducing the computational complexity of conventional adaptive algorithms, mainly for applications which require a large numb of adaptive coefficients. In this paper, we present a nonuniform subband structure with critical sampling, which is able of exact modeling an arbitrary FIR system. We also derive an LMS-type adaptation algorithm with normalized step-sizes, which works the lowest downsampling rate and minimizes the average of the subband squared-errors. A convergence analysis of the propos adaptation algorithm is presented, from which its convergence rate can be estimated

Filter bank design for a subband adaptive filtering structure with critical sampling

Subband adaptive filtering structures are attractive in applications such as acoustic echo cancellation and channel equalization, due to their properties of decorrelating the input signal and reducing the computational complexity. Recently, a new adaptive filtering structure with critical sampling was proposed. In this paper, we describe an optimization procedure to select the analysis and synthesis filter banks of this new subband structure, so that minimum steady-state mean square error or fastest convergence rate can be achieved. Such filter-bank design method is based on a theoretical analysis of the convergence properties of the adaptation algorithm and uses a nonlinear optimization routine. Computer simulations illustrate the convergence improvements that can be obtained with the filter banks designed by the proposed method.

Filter bank design for an adaptive subband structure with critical sampling using a new adaptation scheme

In adaptive filtering, new structures and algorithms are frequently proposed with the objective of increasing the convergence speed and/or reducing the computational complexity of the conventional algorithms, mainly for applications that require a large number of adaptive coefficients. In such applications, adaptive subband structures are specially attractive since the adaptation and filtering are performed at a reduced sampling rate. Recently, an adaptive filtering structure with critical sampling was proposed, which is able to model exactly any finite impulse response system. In this paper a convergence analysis of an LMS type adaptation algorithm with power normalized step-sizes developed for this subband structure is presented and an optimization procedure to select the filter bank coefficients is proposed, which results in fast adaptation convergence rate and reduced mean-square error.

Projeto de banco de filtros para estruturas adaptativas em subbandas

In adaptive filtering, new structures and algorithms are frequently proposed with the objective of increasing the convergence speed and/or reducing the computational complexity of the conventional algorithms, mainly for applications which require a large number of adaptive coefficients. In such applications, adaptive subband structures are especially atractive since the adaptation and filtering are performed at a reduced sampling rate. Recently, new adaptive subband structures were proposed, which are able of modeling exactly any finite impulse response system. In this work the performance properties of two of these structures are investigated: in the first one, composed of an analysis filter bank and sparse subfilters, there is no sampling rate alteration, while in the second one, the ouput signals of the analysis filter bank are maximally decimated. Optimization procedures are proposed to select the filter bank coefficients of such structures, which result in a reduction of the mean-square error (MSE) and/or in an improvement of the convergence behavior of the adaptation algorithms.

Using reverberation time estimates in blind separation of acoustic sources

Blind separation techniques of sound sources, designed to work with voice signals, present a performance highly dependent on the number of coefficients of the separation system. In general, different environments require different lengths of separation filters. This paper proposes the use of reverberation time information arising from lateral blind estimation techniques for tuning the degree of freedom of the separation system in order to blindly obtain a reasonable source-to-distortion ratio (SDR). This tuning enables, for example, that the complexity of the separation system be adjusted in each band if a subband structure is employed for sources separation.

Subband blind source separation considering acoustic reverberation characteristics

Subband blind source separation methods have been recently proposed with the objective of reducing the computational complexity and improving the convergence rate of the online algorithms. Oversampled subband structures with DFT filter banks are usually employed in order to avoid aliasing effects and keep enough samples to estimate the statistics of the subband signals. In this paper we present a critically sampled subband structure, composed of cosine-modulated filter banks and reduced-order adaptive subfilters, for convolutive blind source separation. Its performance is compared to those of the fullband algorithm, of an oversampled subband algorithm and of a frequency-domain algorithm. We also evaluate, through computer simulations, the impact of reducing the order of the high-frequency subfilters on the separation results for different reverberation characteristics.

Performance evaluation of two semi-blind source separation methods

Blind source separation methods resort to very weak hypothesis concerning the source signals, as well as the mixing matrix. This paper verifies experimentally the performance improvement in two different source separation algorithms when some statistical knowledge about the mixing matrix is used. A natural way of inserting such information in source separation methods is to put them in a Bayesian framework. This approach presents immediate applications in digital communication and speech signal processing systems, among many others.

Mean-Square Error Analysis of a Non-Uniform Subband Adaptive Filtering Structure with Critical Sampling

Adaptive subband structures have been proposed with the objective of increasing the convergence speed and/or reducing the computational complexity of conventional adaptive algorithms, mainly for applications which require a large number of adaptive coefficients. In this paper, we present a non-uniform subband structure with critical sampling, which is able of modeling an arbitrary FIR system with reduced aliasing. An LMS-type adaptation algorithm with normalized step-sizes, which works at the lowest downsampling rate and minimizes the average of the subband squared-errors, is derived for the proposed structure. A convergence analysis of the adaptation algorithm is presented, from which the steady-state mean-square error can be estimated.