Showing papers by "Anthony G. Constantinides published in 1997"

A fast recursive shortest spanning tree for image segmentation and edge detection

Abstract: This correspondence presents a fast recursive shortest spanning tree algorithm for image segmentation and edge detection. The conventional algorithm requires a complexity of o(n/sup 2/) for an image of n pixels, while the complexity of our approach is bounded by O(n), which is a new lower bound for algorithms of this kind. The total memory requirement of our fast algorithm is 20% smaller.

New normalized constant modulus algorithms with relaxation

Abstract: A new normalized constant modulus algorithm is proposed that has a more desirable error performance surface (EPS) than the existing constant modulus blind equalization algorithms. We show that for an autoregressive channel, a well-known class of undesirable local solutions (ULSs) does not exist. The EPSs and convergence of the parameters are shown for a number of channels for which well-known algorithms are known to possess ULSs.

Residual echo signal in critically sampled subband acoustic echo cancellers based on IIR and FIR filter banks

Abstract: The residual echo signal characteristics of critically sampled subband acoustic echo cancellers are analyzed. For finite impulse response (FIR) filter banks, the residual echo signal usually has a relatively broad spectral nature around the subband edges. The residual echo signal of power symmetric infinite impulse response (PS-IIR) filter banks, on the other hand, has very narrowband spectral components around the subband edges. These components can be efficiently removed with PS-IIR notch filters that integrate neatly into the filter banks without introducing perceptually noticeable degradation to the near-end speech. This solution has very low computational complexity and does not impinge on the system performance. Simulation studies with recordings from the cockpit of a car, based on a fast QR least-squares adaptive algorithm, demonstrate the potential of this approach for a practical AEC system.

Underdetermined-order recursive least-squares adaptive filtering: the concept and algorithms

Abstract: Underdetermined recursive least-squares (URLS) adaptive filtering is introduced. In particular, the URLS algorithm is derived and shown to be a direct consequence of the principle of minimal disturbance. By exploiting the Hankel structure of the filter input matrix, the fast transversal filter (FTF) version of the URLS algorithm (URLS-FTF) is derived including sliding window and growing window types. The computational complexity is reduced to O(N)+O(m), where N is the adaptive filter length, and m is the order of the URLS algorithm. In addition, the efficient URLS (EURLS) algorithm, which does not compute the filter coefficients explicitly, thereby significantly reducing the computational load, is presented. Some earlier adaptive algorithms such as the averaged LMS, filtered-X LMS, and fast conjugate gradient are shown to be suboptimal approximations of the URLS algorithm. Instrumental variable approximations are also discussed. The URLS algorithm has a whitening effect on the input, signal, which provides immunity to the eigenvalue spread of the input signal correlation matrix. Although the algorithm is sensitive to observation noise, it has good tracking characteristics, and tradeoffs can be found by tuning the step size. The utility of the URLS algorithms, in its basic form and FTF realization, depends heavily on the practical applicability of the mth-order sliding window estimate of the covariance matrix and mth-order PTF relations. The feasibility of the URLS family in practical applications is demonstrated in channel equalization and acoustic echo cancellation.

An 8/spl times/8-block based motion estimation using Kalman filter

Abstract: It is now quite common in the pel-recursive approaches for motion estimation, to find applications of the Kalman filtering technique both in time and frequency domains. In the block-based approach, very few approaches are available of this technique to refine the estimation of motion vectors resulting from fast algorithms such as the three step on a 16/spl times/16-block basis. This paper proposes an 8/spl times/8-block based motion estimation which uses the Kalman filtering technique to improve the motion estimates resulting from both the three step algorithm and the previous 16/spl times/16-block based Kalman application of Kuo et al. (1996). The state-space representation uses a first order auto-regressive model. Comparative results obtained for different classes of video sequences are presented.

Image coding using data-dependent triangulation

Abstract: An image coding technique that combines irregular subsampling and triangulation is proposed. First, a Delaunay triangulation is constructed over a set of visually significant sampling points selected incrementally as the triangulation progresses. Then a data-dependent triangulation is generated, starting from the Delaunay triangulation, by a series of edge swaps based on the quality of the approximation. Bilinear interpolation is used to approximate the triangular regions. The simulation results show the superiority of the proposed algorithm over JPEG at low bit rates (0.206 bpp and 0.113 bpp) as well as its robustness.

Blind adaptive Volterra system identification using barrier function methods for constrained optimisation

Abstract: This paper forms a part of a series of studies we have undertaken, in which the problem of adaptive Volterra system identification is examined. We assume that we have an observed "output" signal derived from a Volterra filter that is driven by a Gaussian input. Both the filter parameters and the input signal are unknown and therefore the problem can be classified as "blind" or unsupervised in nature. In the statistical approach to the solution of the above problem we formulate equations that relate the unknown parameters of the Volterra model with the statistical parameters of the "output" signal. These equations are highly nonlinear and their solution is achieved through a constrained optimisation formulation. We attempt to interpret the constrained problem as an unconstrained one by incorporating various types of the so called "barrier functions". The barrier function (or interior point) methods are believed to have properties which are theoretically or computationally desirable for constrained optimisation. The results of the entire modelling scheme are compared with other contributions.

Constant modulus blind equalisation algorithms under soft constraint satisfaction

Abstract: New constant modulus (CM) algorithms are presented that are based on soft constraint satisfaction. The stationary points of an algorithm in this family are studied for an AR(p) channel and it is shown that Ding-type undesirable local solutions (ULS) do not exist. This is due to the normalisation of the gradient vector and the soft nonlinearity used in these algorithms. Error performance surfaces (EPS) and convergence trajectories from arbitrary initialisations are presented for various channels that support the analytical findings.

Statistical analysis of a mixed-layer x-ray diffraction peak

Abstract: A mathematical model to describe the line shape of an x-ray diffraction peak from stacks of different layers such as, for instance, an interstratified clay mineral has been evolved. The aim was to be able to analyse the proportions of different specific stacking sequences in two-component interstratified samples. A maximum-entropy algorithm was applied to observed powder-diffraction intensities in order to obtain the probability of each stacking sequence. Application to natural smectite - illite clays gave reasonable results.

Guest Editors' Introduction: Neural Networks For Signal Processing

Abstract: Guest Editors’ Introduction: Neural Networks for Signal Processing PROGRESS in the theory and design of neural networks has expanded on many fronts during the past ten years. Much of that progress has a direct bearing on signal processing. In particular, the nonlinear nature of a neuron that constitutes the basic building block of neural networks—the ability of neural networks to learn from their environments in supervised as well as unsupervised ways and the universal approximation property of neural networks—make them highly suited for solving difficult signal processing problems. Applications of neural networks include • nonlinear signal modeling, detection and estimation, and pattern classification; • system identification, adaptive filtering, and blind adaptation; • image and speech processing; • unconventional applications. From a signal processing perspective, it is imperative that we develop proper understanding of neural networkbased signal processing algorithms and how they impact the above-mentioned applications. We need to assess the impact of neural networks on the performance, robustness, and cost-effectiveness of signal processing systems and develop methodologies for integrating neural networks with other signal processing algorithms. Another important issue is how to evaluate neural network paradigms, learning algorithms, neural network structures, and identify those that work and those that do not work reliably for solving signal processing problems. The issues raised herein demand detailed attention, innovative ways of thinking, and above all, honest answers. This special issue offers a unique forum for researchers and practitioners in this field to present their views on these important questions. The response to the initial call for papers was overwhelming—a total of 101 manuscripts were submitted from all over the world. All guest editors worked hard to keep up with the schedule while maintaining the same high standard as that applied to other manuscripts submitted to the IEEE TRANSACTIONS ON SIGNAL PROCESSING. Each manuscript was reviewed by two to four anonymous reviewers, and then, the reviewers’ recommendation, as well as the manuscript itself, were examined by two guest editors before a final decision was made. Based on the subject area of interest, the accepted papers are grouped into the following categories: • nonlinear signal learning and processing: theory and algorithms; • signal prediction and filtering; • blind source separation and channel equalization; • pattern classification. The first group of papers, which contain five papers and two correspondence items, concerns the theory and algorithms

Block momentum-LMS algorithm based on the method of parallel tangents

Abstract: Based on the method of parallel tangents, the block-LMS algorithm is modified, and the block momentum-LMS algorithm is proposed. The new algorithm has lower computational complexity than the LMS algorithm. It converges significantly faster than the block-LMS algorithm when the input signal is coloured. The time-constant, mean and mean-square convergence conditions and the misadjustment of the proposed algorithm are derived. As a special case, an accurate mean-square convergence condition is obtained for the block-LMS algorithm. Extension to the frequency domain is also discussed. Comprehensive experimental results on system identification and channel equalisation are presented that validate the theoretical findings.

Nonlinear adaptive prediction of nonstationary signals with application to speech coding

Abstract: The purpose of this contribution is to present a new approach for the prediction of speech signals that is appropriate to speech coding. The procedure is based upon the principles of blind equalisation. In an earlier publication we examined these principles from the prediction point of view as a general method. The present contribution examines the approach in relation to speech signal representation for coding and compression. The method outlined in this contribution offers significant advantages compared to the standard prediction methods. This is because the signal estimation is carried out on a sample by sample basis, it needs no estimation of the covariance matrix or some other long term statistical attributes, it makes no assumption on a minimum phase vocal tract transfer function, and hence, it is faithful to the nonstationarities in the analysed signal. The method can be seen as mapping a given signal into a set of signals of increased correlational properties, which in turn may be so mapped until the signals exhibit piecewise constant behaviour. At this stage they are easily modelled and the reverse process can be put into effect for the original signal reconstruction. Speech signals can be rendered piecewise constant approximately after some signal decompositions. Then, the percentage error of predicted value is involved to examine if the decomposition stops. Examples illustrating these principles are included.

A digital signal processing perspective on a stability test of linear system transfer functions

Abstract: The stability of a linear transfer function is fundamental in its real time realisation. Several tests have been developed in the past to test whether a given transfer function is stable. Invariably these tests rely on either a direct use of a Hurwitz test or some convenient modification of it such as the Schur-Cohen test or the Jury-Marden test. A new test is proposed in this contribution which does not rely on the same principles but on the Cauchy integral, and its implementation in practice requires merely the use of the Fast Fourier Transform. The test is convenient for real time implementation.

Mitigating array sensor malfunctions with mixed second and fourth order statistics

Abstract: The authors propose a modified MUSIC algorithm which employs mixed second and fourth order statistics. When an array suffers from sensor malfunction, higher order statistics are used to extend the effective aperture of the faulty array. Simulation results support the potential of this approach which maximises the additional information available in fourth order statistics, while retaining the statistical reliability of second order sample estimates.

Optimization for sigma-delta modulators

Abstract: The performance of a higher-order single-stage sigma-delta modulator (/spl Sigma//spl Delta/M) is determined by the loop filter in which the quantizer is embedded. Not the coefficients in the loop filter themselves, but ratios of them determine the input-output relationship in the modulator and thus its performance. The ratios of coefficients that affect the performance of the /spl Sigma//spl Delta/M are found, thus establishing a search space for optimization of the modulators. An optimization method of the random search type is presented and used for optimizing the /spl Sigma//spl Delta/M to any particular input.

A cumulant based variable step size adaptive algorithm

Abstract: An adaptive step size selection rule for least mean squares (LMS) adaptive algorithms is introduced. The step size sequence is adjusted using the kurtosis of the estimation error, thus the performance degradation due to the existence of noise with a strong variance. The proposed algorithm traces changes in the noise statistics and optimally adapts itself, exhibiting a reduced steady state error. Simulation results illustrate the algorithm's superior performance and confirm its ability to optimally adapt to time varying noise environments. The convergence behavior of the algorithm is also addressed.