In applications involving system identification problems, some characteristics of the impulse response of the system to be identified are usually exploited to design adaptive algorithms with improved performance. In this context, this paper focuses on the identification of systems that own intrinsic symmetric or antisymmetric properties, which can be further formulated by using a combination of bilinear forms. Based on such an approach, the least-mean-square (LMS) and normalized LMS (NLMS) algorithms with symmetric/antisymmetric properties (termed here LMS-SAS and NLMS-SAS) are proposed. Simulation results are shown confirming the improved convergence speed achieved by the proposed algorithms as compared to the conventional LMS and NLMS counterparts for different operating scenarios.

This letter presents a time-varying autoregressive (TVAR) model aiming to characterize nonstationary behaviors often observed in real-world processes, which cannot be properly described by autoregressive processes such as first-order Markov and random-walk models. Specifically, general model expressions for the mean vector and covariance matrix of the TVAR model are firstly derived. Then, such expressions are used to guide the design of two special setups for the TVAR model. The capability of the developed model to reproduce important nonstationary behaviors is verified mathematically and through simulations.

This letter presents model expressions describing the steady-state behavior of proportionate normalized least-mean-square (PNLMS)-type algorithms, taking into account both complex- and real-valued correlated Gaussian input data. Specifically, based on energy-conservation arguments, general expressions for the excess mean-square error (EMSE) in steady state and misadjustment are obtained. Such general expressions are then applied to two well-known PNLMS-type algorithms, namely the improved PNLMS (IPNLMS) and the individual-activation-factor PNLMS (IAF-PNLMS). Simulation results are shown confirming the accuracy of the proposed model expressions under different operating conditions.

The increasing demand for mobile communications has led to a spectrum shortage that severely limits the expansion of cellular systems. An attractive approach to cope with this problem is to use beamforming techniques for improving the signal-to-interference-plus-noise ratio (SINR), which in turn allows reducing the transmission power and increasing the system capacity. In this context, a novel approach for joint beamforming and power control in cellular systems is discussed. Regarding the beamforming, an adaptive algorithm is devised, aiming to overcome an important implementation challenge of the constrained stochastic gradient (CSG) and improved CSG (ICSG) algorithms, namely the requirement to estimate each interfering signal individually. The proposed beamforming algorithm, called adaptive-projection CSG (AP-CSG), provides both enhanced SINR performance and reduced computational burden as compared with the standard CSG algorithms. Concerning the power control, a real-time algorithm designed to operate jointly with the AP-CSG is also developed, which allows reducing the transmission power while maintaining acceptable SINR levels. Moreover, a stochastic model for the proposed joint beamforming and power control algorithm is derived, allowing the prediction of its behavior accurately. Numerical simulation results are shown, confirming the effectiveness of the new algorithm and the accuracy of the proposed model.

The motion estimation process and coding mode selection are responsible for a large portion of the computational effort in H.264-based video encoding systems optimized for rate-distortion (RD). This paper presents a rate sorting and truncation strategy that incorporates the RD optimization criterion in the decision to evaluate distortion for both motion vectors and coding modes. Experimental results confirm the effectiveness of the proposed approach, yielding up to a 90% reduction in the computational complexity. An additional saving can also be obtained, with insignificant RD performance loss, by using a quality threshold.

This paper addresses the use of full reference metrics in the characterization of perceptual difference in a pair of images, in which one of them is degraded by additive white Gaussian noise (AWGN). Conventional (SNR, PSNR, and CC) and perceptual (MSSIM, IFC, VIF, and C4) full reference metrics are considered. The subjective experiment, which upholds the presented results, has been carried out with approximately 120 human subjects, considering a test set of 60 images divided into 5 groups. The difference detection threshold is defined as the value for a given metric from which human subjects perceive the difference between the reference and test images. In this sense, the performed data analysis aims both to derive difference detection thresholds for each tested metric and to evaluate their consistency.

The major drawback for using adaptive Volterra filters is the high computational complexity requirement. In this context, a large number of reduced complexity implementations have been proposed to increase the applicability of such filters. Contributing in this sense, this paper presents a fully LMS adaptive approach for implementing interpolated Volterra filters with a very good performance characteristic. The adaptive interpolated Volterra structure is a simplified version of the conventional one which adapts both the interpolator and the sparse filter. Numerical simulations illustrate the usefulness of the proposed approach.

This paper presents a procedure for implementing fully adaptive interpolated FIR filters with removed border effect. The proposed approach allows reducing the steady-state mean-square error by eliminating the main sources of performance degradation from the adaptive interpolated FIR filters. In addition, the computational effort needed for implementing such a procedure is very small. Simulation results confirm the effectiveness of the proposed approach.

This paper presents a procedure to remove the border effect from interpolated finite impulse response (IFIR) and interpolated Volterra adaptive filters using the LMS algorithm. The used approach permits to reduce the steady-state mean-square error (MSE) of such structures. In situations that the border effect is important, the obtained improvement is noticeable. In addition, the computational burden required to implement such a procedure is slight. Simulation results confirm the effectiveness of the proposed approach.

In this work, a stochastic model for the mean weight behavior and the learning curve of the noise constrained least-mean-square (NCLMS) algorithm is presented. The proposed model is simpler than that recently presented in the open literature. The main feature of this algorithm is that it takes into account the additive noise variance in the mean-square error (MSE) minimization process. As a result, some additional control parameters are included in the adaptive algorithm, affecting the convergence behavior of the algorithm. Then, some hints regarding these parameter settings for algorithm stability are also given. Through numerical simulations the accuracy of the proposed model is confirmed.

This paper discusses the implementation of holographic associative memories through artificial neural networks (using mirror neurons). Such memories use a data structure based on random patterns to store and retrieve the desired information, presenting some features that are very similar to biological memory systems. This approach for data storage discards any training process as well as does not need complex procedures to retrieve the output data. As the information is distributed in the whole memory, some part of it may be corrupted by noise or even be deleted without any loss of essential information. An application example is shown and discussed.

This paper proposes a stochastic model for the generalized subband decomposition normalized LMS (NLMS) algorithm. This algorithm is used as an alternative to the standard NLMS one, aiming to improve the convergence speed under correlated input data. Analytical models for the first and second moments of the filter weights are derived taking into account the time-varying nature of normalized step size. Moreover, in the model expressions a positive regularization parameter ε is added to the power estimates, preventing division by zero during the power normalization process. Through simulation results, the accuracy of the proposed analytical model can be verified.

This paper proposes a new stochastic model for the normalized LMS (NLMS) algorithm under correlated input data. The proposed model is derived without invoking the simplifying assumption that xT(n)x(n) has a chi-square distribution to determine E{1/[xT(n)x(n)/N]}. Under correlated input data that assumption is not correct and thus the resulting model becomes inaccurate. Without considering such simplifying assumption, a high-order hyperelliptic integral has to be computed. The proposed model is based on tackling the solution of that integral. Numerical simulations verify the quality of the proposed model.

In this paper three image quality metrics are compared: mean-square error (MSE), noise quality measure (NQM), and structural information metric (SIM). Such a comparison is made in order to evaluate the performance of those different metrics under practical conditions. Experimental results along with statistical indices of performance are provided. The referred results suggest that there are situations in which the MSE-based metric outperforms the two other considered metrics.

This paper presents a stochastic analysis of the leaky filtered-X least-mean-square (LFXLMS) algorithm. The version with leakage of the adaptive algorithm is used in practical implementations aiming to reduce undesirable effects due to numerical errors in finite-precision machines, overload of the secondary source, among others. Based on new analysis assumptions, instead of the ordinary independence theory frequently used in classical LMS analysis, an analytical model for the first and second moments of the adaptive filter weights has been derived. In addition, the proposed theoretical models consider the situation in which the secondary path is imperfectly modeled. Experimental results demonstrate the accuracy of the proposed model as compared with the classical analysis.

This paper presents a new motion estimation (ME) strategy for video coding. Such a strategy makes use of stopping rules in the ME process, which considers both the motion vector coding cost and desired minimum quality for each macroblock used. In a rate-distortion sense, the obtained results by the proposed algorithm are superior to the ones achieved with the full search (FS) algorithm. Moreover, the strategy reduces the mean number of searching points. Through simulations, by using an H.263 encoder, we verify that the proposed algorithm is more effective for video coding than the one used in the TMN11-LC (low complexity mode decision) FS implementation.

This paper presents an improved embedded zerotree wavelet (EZW) coding algorithm, which makes use of the wavelet regularity to derive a classification criterion of wavelet coefficients in spatial-orientation hierarchical trees. Variations of the EZW algorithm discussed in the open literature have proposed some modifications in the process of exploiting the similarity of coefficients through the scales, however, not defining a figure of merit to measure such a similarity. Simulation results achieved from the coding of well-known images in the literature, for several bit rates, show a better performance of the proposed algorithm in both PSNR and subjective terms, as compared with EZW and SPIHT algorithms.

The paper addresses the issue of perceptual image quality assessment. By using Bayesian networks, we propose a Bayesian composed quality measure (B-CQM). This metric can assess quality in images degraded by combined noise injection and frequency distortion. It presents some advantages with respect to the original CQM approach, such as upholding the stochastic nature of the subjective quality assessment and easier inclusion of the effect of new experimental data in the metric model by just updating its probability tables. Some examples are provided in order to verify the behavior of the proposed metric.

This paper addresses the issue of perceptual image quality assessment in image restoration systems. Firstly, experiments are conducted to evaluate perceived quality in images degraded only by frequency distortion. Based on the resulting experimental data, a distortion quality measure (DQM) is proposed. Then another experiment, whose test images present frequency distortion and noise injection, is achieved. From this latter experiment, a composed quality measure (CQM) is developed to assess perceived quality of images degraded by combined effects of frequency distortion and noise injection. The CQM is derived from DQM and NQM (noise quality measure), this latter has been recently proposed, and it can be used as a tool for evaluation and optimization of image restoration systems according to human visual perception.

This paper presents a stochastic analysis of the Delayed LMS adaptive algorithm with leakage. Such an analysis is derived without invoking the independence assumption, giving rise to a new model. It permits explicitly to take into account mismatches between the system delay and its estimate. This feature is not available in previous models. In addition, through the introduction of a leakage factor, we can keep the adaptive algorithm stability under imperfect delay estimate. Recursive difference equations are derived for the first and second moments of the weight vector. Simulation results are presented to support the proposed model.

Methods for histogram thresholding based on the minimization of a threshold-dependent criterion function might not work well for images having multimodal histograms. We propose an approach to threshold the histogram according to the similarity between gray levels. Such a similarity is assessed through a fuzzy measure. In this way, we overcome the local minima that affect most of the conventional methods. The experimental results demonstrate the effectiveness of the proposed approach for both bimodal and multimodal histograms.

The image restoration via reduced order model Kalman filter (ROMKF) is accomplished in conjunction with a maximum likelihood technique for image/blur parameter estimation purposes. Traditionally, one uses initial condition sensitive optimization algorithms at the estimation stage. This work concerns the use of evolutionary programming (EP) in the parameter estimation phase of the ROMKF space-adaptive image restoration. Experimental comparisons between both of the mentioned optimization strategies are presented. Simulation results suggest that more reliable ROMKF restorations are obtained when less initial condition sensitive algorithms are adopted.

This paper proposes a design methodology for transitional low-pass filters using six classical approximations of polynomial filters. The design is achieved taking into account a prescribed specification, leading to a better trade-off among the magnitude, phase and/or time responses. With this approach it is possible to design filters that have an improved performance than those designed with classical polynomial approximations or some other transitional filters proposed in the literature. An example demonstrating the results and effectiveness of our proposal is presented.

This paper presents an analytical model for the mean weight behavior of two AIFIR structures using the LMS algorithm to adapt the sparse filter weights. The introduction of an interpolating block cascaded with the adaptive sparse filter imposes signal correlations. On the other hand, it is well known that such correlations are disregarded by the independence theory, which is the base for the stochastic analysis of the LMS algorithm. The proposed models have been derived without using the independence theory. Simulation results demonstrate the effectiveness of the proposed analytical models as compared with the classical analysis.

This paper presents a new analytical model for the second moment behavior of the filtered-X LMS algorithm. The new model is not based on the independence theory, and is derived for Gaussian inputs and slow adaptation. Monte Carlo simulations show excellent agreement with the behavior predicted by the theoretical model.

This paper presents a stochastic analysis of the Filtered-X LMS algorithm. The mean weight vector recursion is derived for slow adaptation and for a white reference signal without use of independence theory. The Wiener solution is determined explicitly as a function of the input statistics and the impulse responses of the primary and secondary signal paths. It is shown that the steady-state mean weights for the Filtered-X LMS algorithm converge to the Wiener solution only if the estimate of the secondary path is without error. Monte Carlo simulations show excellent agreement with the behavior predicted by the theoretical model.