The ZA-NLMS (for zero-attractor) represents arguably the seminal sparsity-aware gradient adaptive algorithm. As it is constraint by the ℓ1-norm of the filter weights, the underlying problem turns convex, hence with unique solution (in expected sense). Despite these friendly properties, the algorithm convergence and, more important, the best-performing sparsity tradeoff are yet to be effectively studied. This paper presents a comprehensive analytical study on ZA-NLMS' convergence, which results in the optimal (constant) sparsity tradeoff. The value of this decisive hyperparameter from a practitioner point of view turns out related to the 3/2-power of the adaptive filter length. This outcome, difficult to argue intuitively, as well as the convergence model, have been exhaustively validated with numerical experiments.

Estimating vehicles' speed is of great importance for the vehicular technology society (VTS) and the intelligent transportation society (ITS) for the next generation of self-driven smart cars. Having an accurate estimation of vehicle speed is critical for safe driving, collision avoidance, driver assistance systems, etc. This paper proposes a novel method intended for vehicle speed estimation from a camera in motion. Specifically, the system utilizes the smart phone camera mounted behind the windshield using a phone holder to capture consecutive frames of the street ahead, then it accurately estimates the distance, in centimeters accuracy, of the frontal vehicles and their speeds, while the vehicle with the smartphone is in motion. Initial testing for the system shows very promising results with a worst- case accuracy of 96%.

With the fast paced improvements in smart devices such as tablets and smartphones, many schools in the developed countries are trying to improve the traditional pedagogical methods by harnessing smart learning technology for different purposes. In this paper, we proposed the use of a smart novel educational technology to immerse the students in the learning process by engaging as many of the students' senses as possible through the creation of interactive elements and entertaining features. This process is also called edutainment, combining both education and entertainment, and it works by making students learn subconsciously in an immersive manner. The aim of this paper is to demonstrate how an immersive augmented reality application can work in conjunction with Arabic schoolbooks in an attempt to improve Arabic literacy for children in the UAE. The reason for this is that children in the UAE are being exposed to English most of the time and, as a consequence, gradually losing their mother tongue. A focus field-study was then conducted on two groups of Grade 1 students to get a pre-understanding in the impact of such educational technology, targeted for younger learners, in language literacy education.

Embedding the norm in gradient-based adaptive filtering is a popular solution for sparse plant estimation. Even though the foundations are well understood, the selection of the sparsity hyper-parameter still remains today matter of study. Supported on the modal analysis of the adaptive algorithm near steady state, this paper shows that the optimal sparsity tradeoff depends on filter length, plant sparsity and signal-to-noise ratio. In a practical implementation, these terms are obtained with an unsupervised mechanism tracking the filter weights. Simulation results prove the robustness and superiority of the novel adaptive-tradeoff sparsity-aware method.

The ℓ0-normalized least mean squares (ℓ0-NLMS) is arguably the reference gradient adaptive algorithm for sparse system estimation. However, alike all sparse gradient adaptive algorithms, the ℓ0-NLMS performance is sensitive to the (adequate) selection of the tradeoff parameter. Highlighted in this letter, the existence of two convergence modes, linked to the negligible and to the significant taps, paves the way for the convergence analysis, which results in a set of nonlinear (quadratic) convergence equations. Therefrom, the minimization of the steady-state misalignment concludes in the optimal tradeoff, which happens to relate to the NLMS step size, filter length, plant sparsity, and noise level in an extremely compact fashion. Exhaustive simulation experiments show strong agreement between the analytical predictions and the empirical performance.

In order to exploit the inherent sparsity of an OFDM multipath wireless channel, this paper presents a novel estimation technique based on the ℓ1-LS criterion that relies on the selection of the optimal regularization parameter. This term is proven in the paper to be intimately related to the Signal to Noise Ratio (SNR) and Sparsity Degree (ρ) of the channel. Therefore, in order to take full advantage of the ℓ1-LS framework, an ad-hoc technique for estimating blindly the SNR and ρ of the problem is also introduced. The performance of the novel algorithm is compared to a number of reference techniques. Two standard measures, namely the Mean Square Error (MSE) of the estimated channel and the Symbol Error Rate (SER) of the OFDM system validate the robustness of the proposed algorithm. Indeed, this novel technique achieves better estimation along with better spectrum efficiency.

Embedding a sparse penalty in conventional Least Mean Square (LMS) adaptive algorithms is an established strategy to enhance the performance and robustness against noise in the estimation of sparse plants, such as wireless mul-tipath channels. In this paper we review the most prominent NLMS-based algorithms with ℓp-norm constraint, discussing the underlying mechanisms that lead to improvement gains in sparse scenarios. Simulation results validate the analysis and comparative discussion. Given that adaptive algorithms operating in time domain deteriorate with correlated signals, we propose hereby a novel frequency-domain (FD) ℓp-NLMS that performs in such situations. Simulation results indicate that the proposed method outperforms its time-domain counterparts not only in convergence rate but more importantly in residual misalignment. This important result has not been echoed so far.

This paper presents a fast approach for detecting coherent signals encountered in mul-tipath propagation environment. The developed approach utilizes a Toeplitz matrix reconstruction method along with DOA estimation using Root-MUSIC. The proposed approach exhibits several features such as robustness, high accuracy, and computational efficiency. Nevertheless, the primary advantage of the proposed Toeplitz-based de-correlation scheme is the fast detection of the maximum possible number of coherent signals, without the need of iteration and re-estimation.

This letter presents the exact normalized least-mean-square (NLMS) algorithm for the lp-norm-regularized square error, a popular choice for the identification of sparse systems corrupted by additive noise. The resulting exact lp-NLMS algorithm manifests differences to the original one, such as an independent update for each weight, a new sparsity-promoting compensated update, and the guarantee of stable convergence for any configuration (regardless the choice of lp norm and sparsity-tradeoff constant). Simulation results show that the exact lp-NLMS is stable and it outperforms the original one, thus validating the optimality of the proposed methodology.

Fitting a multivariate Gaussian mixture to data represents an attractive, as well as challenging problem, in especial when sparsity in the solution is demanded. Achieving this objective requires the concurrent update of all parameters (weight, centers, and precisions) of all multivariate Gaussian functions during the learning process. Such is the focus of this paper, which presents a novel method founded on the minimization of the error of the generalized logarithmic utility function (GLUF). This choice, which allows us to move smoothly from the mean square error (MSE) criterion to the one based on the logarithmic error, yields an optimization problem that resembles a locally convex problem and can be solved with a quasi-Newton method. The GLUF framework also facilitates the comparative study between both extremes, concluding that the classical MSE optimization is not the most adequate for the task. The performance of the proposed novel technique is demonstrated on simulated as well as realistic scenarios.

This paper presents a novel method for the compensation of near-end interferences in acoustic echo cancellation. The technique is derived from a rigorous analysis dealing with the statistical impact of the update over the “look-ahead” error. The outcome is a well-conditioned problem, whose numerical solution is shaped from maximum-likelihood principles. Results on a realistic simulated scenario show the excellent performance and robustness of the proposed method.

This paper presents a novel method for fitting a sparse Gaussian mixture on a non-negative function of reference. Despite this problem is well known to be highly non-linear, the use of the logarithmic utility function seems to alleviate such an undesired situation. Moreover, sparsity measures related to the Gaussian precision can be integrated in such a way that the numerical algorithm turns out easy to implement, it is numerically efficient, and presents stable convergence. The method has been originally thought for the parameterization of power spectra, but it can also be useful in different scenarios.

The effects of noise in autoregressive (AR) analysis (or linear prediction) and its compensation (NCAR) has been commonly carried out in the time domain under the least square (LS) criterion. This paper studies the adequacy of such an approach by means of a comparative analysis with selected frequency-based NCAR methods. In particular, the maximization of the spectral likelihood (ML) results in a proper optimization problem that is easy to solve and brings useful insights into the rationale of the NCAR problem. On the contrary, popular time-based NCAR methods are shown in the paper to be designed, in the ML context, around ill-conditioned criteria, requiring constraints to guarantee stable solutions. The statistical analysis on a realistic scenario as well as an experiment on speech enhancement complement this analysis.

The problem of noise-compensated autoregressive estimation has not been sufficiently explored especially with regard to the variance of the estimation. This paper explores this important aspect, presenting the asymptotic Cramer-Rao bound thereto. This valuable result is achieved by using a frequency- domain perspective of the problem as well as an unusual parametrization of an autoregressive model. One interesting finding is that the Fisher information matrix turns out to be built with the Wiener filter rule. Despite the power spectral density of the noise is assumed to be available in advance, the variance of the best estimator thereto is proven to be larger than that of the classical (noiseless) autoregressive estimation. The theoretical analysis has been validated with simulation experiments involving stationary colored noise.

This paper proposes a novel method for autoregressive estimation on additive noise. The method is founded on the maximum-likelihood criterion in the spectral domain. This problem statement yields a non-linear optimization problem that can be revamped as a re-weighted least square problem. The resulting spectral weighting function turns out to be an integer power of the Wiener filter, this meaning that spectral regions with higher signal-to-noise ratio are more relevant in the estimation. Furthermore, this frequency-selective scenario allows to interpret this problem as one with missing samples. Simulated experiments prove the validity of the problem statement, showing as well the excellent performance of the proposed algorithm.

Frequency-selective autoregressive (AR) estimation is arousing increasing interest. We propose herein a new method to estimate the AR model from a reduced set of spectral samples. The proposed method is founded on the maximum likelihood criterion over the logarithmic spectral residue, and it is implemented efficiently with a multivariate Newton-Raphson algorithm. Results over deterministic and stochastic scenarios show its excellent performance.

Probably the most powerful method for speech analysis is the linear prediction analysis, or LPC analysis, one of its main characteristics being the estimation of time-domain related parameters from time-domain samples. This paper proposes a novel speech analysis framework for estimating the spectral poles directly from spectral samples in voiced speech utterances. The method can be described in plain words as the task of fitting the spectral envelope of an allpole model directly on the log energy of the harmonics. This problem is addressed with an analysis-by-synthesis mechanism supported on a Newton-Raphson algorithm of fast convergence. The proposed method differs clearly from previous approaches commonly used in Harmonic or Sinusoidal Coding. Comparative results on synthetic signals show the excellent performance of the novel analysis technique.

The architecture of the cerebellar model articulation controller (CMAC) presents a rigid compromise between learning and generalization. In the presence of a sparse training dataset, this limitation manifestly causes overfitting, a drawback that is not overcome by current training algorithms. This paper proposes a novel training framework founded on the Tikhonov regularization, which relates to the minimization of the power of the sigma-order derivative. This smoothness criterion yields to an internal cell-interaction mechanism that increases the generalization beyond the degree hardcoded in the CMAC architecture while preserving the potential CMAC learning capabilities. The resulting training mechanism, which proves to be simple and computationally efficient, is deduced from a rigorous theoretical study. The performance of the new training framework is validated against comparative benchmarks from the DELVE environment.

This paper proposes a novel technique for the time-frequency (TF) estimation of a Gaussian chirplet mixture in a signal. The technique relies on the combination of one and a half-worth of chirp transform of the signal and an expectation maximization-driven blind identification algorithm, in such a way that it carries out the chirplet estimation in-segment. The mechanism resembles stereo processing while keeping an interesting biological parallel with auditory cortex artificial models. This technique applied on a short-time processing basis is shown to be accurate at identifying the TF composition of a bat echolocation chirp. The proposed technique represents the initial basis for the development of a parallel hierarchic processing for more ambitious objectives.

The common approach to image matching is to detect spatial features present in both images and create a mapping that relates both images. The main drawback of this method takes place when more than one matching is likely. A first simplification to this ambiguity is to represent with a parametric model the point locus where the matching is highly likely, and then use a POCS (projection onto convex sets) procedure combined with Tikhonov regularization that results in the mapping vectors. However, if there is more than one model per pixel, the regularization and constraint-forcing process faces a multiple-choice dilemma that has no easy solution. This work proposes a framework to overcome this drawback: the combined projection over multiple models based on the L/sub k/, norm of the projection-point distance. This approach is tested on a stereo-pair that presents multiple choices of similar likelihood.

Artificial muscles are formed by attaching a conducting polymeric film to a non-conducting one. The flow of an electric current produces a macroscopic bending movement on the muscle. A good characterization of both, motion rate and energy of curvature, is required for improving the efficiency of these devices. In this paper, a two-cam stereo vision system is proposed to acquire and process the image sequence and a 3D snake for tracking the muscle. From the curve given by the snake, mechanical parameters of the artificial muscle can be estimated. The movements along the life cycle of the muscle can be compared with the energy consumed in each cycle. This is necessary for determining the span life of these devices in applications where they work as actuators. Results prove the validity of this approach.

This paper introduces a new spectral representation-based pitch estimation method. Since pitch is never stationary during real conversations, but often undergoes changes because of intonation, the spectral representation is derived from the Shorttime Harmonic Chirp Transform. This lets our technique to perform very well in noisy conditions, and to extract pitch values with high confidence, even from segments with strong intonations. The paper discusses a new way of segment-vice pitch extraction and does not deal with continuous pitch tracking, which is a topic of our future work. However, the performance of the proposed method is demonstrated on real recordings and the noise-dependency of its accuracy is numerically analyzed. 1.

The CMAC neural network presents a rigid architecture for learning and generalizing simultaneously, a limitation stressed with sparse or non-dense training datasets, and hardly solved by the current training algorithms. This paper proposes a novel training algorithm that overcomes the mentioned tradeoff. The training mechanism is based on the minimization of the stiffness energy of the output, solution based on the active deformable model theory. These ideas lead to a cell-interaction-based internal update mechanism that preserves the potential CMAC learning capabilities and delivers a higher generalization degree than the one a-priori embedded in the CMAC architecture. The training mechanism is derived entirely from a rigorous theoretical study. This analysis is supported with comparative results on the inverse kinematics of a robotic arm, which prove the excellent performance of the proposed active training.

Active deformable models are simple tools, very popular in computer vision and computer graphics, for solving ill-posed problems or mimic real physical systems. The classical formulation is given in the spatial domain, the motor of the procedure is a second-order linear system, and rigidity and elasticity are the basic parameters for its characterization. This paper proposes a novel formulation based on a frequency-domain analysis: The internal energy functional and the Lagrange minimization are performed entirely in the frequency domain, which leads to a simple formulation and design. The frequency-based implementation offers important computational savings in comparison to the original one, a feature that is improved by the efficient hardware and software computation of the FFT algorithm. This new formulation focuses on the stiffness spectrum, allowing the possibility of constructing deformable models apart from the elasticity and rigidity-based original formulation. Simulation examples validate the theoretical results.

Artificial muscles are formed by attaching a conducting polymeric film to a non-conducting one. Applying an electrical current on the muscle, a macroscopic bending movement appears on it. Study of curvature variations and related parameters, such as speed of motion or energy of curvature, is necessary for improving the efficiency of these devices. In a previous work, a one-cam computer vision system was developed to estimate motion parameters in 2D with precise results. In this paper, a two-cam stereo vision system is proposed to process the image sequence and track the muscle in 3D. Active contours models are employed in motion detection and mechanical parameters estimation. Results prove the validity of this approach, allowing automatic testing on the research into artificial muscles.

In radiotherapy (RT), organ motion caused by breathing prevents accurate patient positioning, radiation dose, and target volume determination. Most of the motion-compensated trial techniques require collaboration of the patient and expensive equipment. Estimating the motion between two computed tomography (CT) three-dimensional scans at the extremes of the breathing cycle and including this information in the RT planning has been shyly considered, mainly because that is a tedious manual task. This paper proposes a method to compute in a fully automatic fashion the spatial correspondence between those sets of volumetric CT data. Given the large ambiguity present in this problem, the method aims to reduce gradually this uncertainty through two main phases: a similarity-parametrization data analysis and a projection-regularization phase. Results on a real study show a high accuracy in establishing the spatial correspondence between both sets. Embedding this method in RT planning tools is foreseen, after making some suggested improvements and proving the validity of the two-scan approach.

An optimal least-mean-square (LMS) algorithm is presented to be used with a lookup-table plus transversal filter adaptive echo canceller that has independent steps for the non-overlapped and overlapped parts of the transversal structure. Its advantage over previous schemes and algorithms used with this family of structures is presented. A suboptimal switched step version is discussed. Some simulation examples illustrate the performances of these schemes.>