If you want to read the PDF, try requesting it from the authors.

Abstract

Nonlinear spline adaptive filters are a class of adaptive filters for modelling nonlinear systems. To improve the convergence performance of existing nonlinear spline adaptive filters (SAFs), in this paper, we propose a low rank approximation for different SAF models by incorporating the technique of nearest Kronecker product decomposition. We consider the Wiener and Hammerstein SAF models for developing the proposed algorithms, and simulation studies carried out show that improved convergence and tracking performance can be achieved compared to traditional SAFs.

No full-text available

Request Full-text Paper PDF

To read the full-text of this research,
you can request a copy directly from the authors.

ResearchGate has not been able to resolve any citations for this publication.
Article
Full-text available
Linear system identification is a key problem in many important applications, among which echo cancellation is a very challenging one. Due to the long length impulse responses (i.e., echo paths) to be identified, there is always room (and needs) to improve the performance of the echo cancellers, especially in terms of complexity, convergence rate, robustness, and accuracy. In this paper, we propose a new way to address the system identification problem (from the echo cancellation perspective), by exploiting an optimal approximation of the impulse response based on the nearest Kronecker product decomposition. Also, we make a first step toward this direction, by developing an iterative Wiener filter based on this approach. As compared to the conventional Wiener filter, the proposed solution is much more attractive, since its gain is twofold. First, the matrices to be inverted (or, preferably, linear systems to be solved) are smaller as compared to the conventional approach. Second, as a consequence, the iterative Wiener filter leads to a good estimate of the impulse response, even when a small amount of data is available for the estimation of the statistics. Simulation results support the theoretical findings and indicate the good results of the proposed approach, for the identification of different network and acoustic impulse responses.
Article
Full-text available
We present a novel diffusion scheme for online kernel-based learning over networks. So far, a major drawback of any online learning algorithm, operating in a reproducing kernel Hilbert space (RKHS), is the need for updating a growing number of parameters as time iterations evolve. Besides complexity, this leads to an increased need of communication resources, in a distributed setting. In contrast, the proposed method approximates the solution as a fixed-size vector (of larger dimension than the input space) using Random Fourier Features. This paves the way to use standard linear combine-then-adapt techniques. To the best of our knowledge, this is the first time that a complete protocol for distributed online learning in RKHS is presented. Conditions for asymptotic convergence and boundness of the networkwise regret are also provided. The simulated tests illustrate the performance of the proposed scheme.
Conference Paper
Full-text available
Recently, a nonlinear acoustic echo cancellation algorithm based on the framework of kernel methods has been proposed by modeling the echo path as a Hammerstein system. However, it requires a large amount of computation to be implemented. In this paper, we propose to use iterative methods for solving linear systems in order to reduce the numerical complexity of identifying the nonlinear and linear parts of the echo path. Also we investigate the effect on performance of the parameters of the methods for both iterative batch and online versions. Simulation results confirm the complexity reduction and good performance of the proposed methods for clipping nonlinearity and room impulse response variation.
Article
Full-text available
In this paper two novel nonlinear cascade adaptive architectures, here called sandwich models, suitable for the identification of general nonlinear systems are presented. The proposed architectures rely on the combination of structural blocks, each one implementing a linear filter or a memoryless nonlinear function. All the nonlinear functions involved in the adaptation process are based on spline functions and can be easily modified during learning using gradient-based techniques. In particular, a simple form of the on-line adaptation algorithms for the two architectures is derived. In addition, we analytically obtain a bound for the selection of the learning rates involved in the learning algorithms, in order to guarantee a convergence towards a minimum of the cost function. Finally, some experimental results demonstrate the effectiveness of the proposed method.
Article
This paper investigates the M-estimate affine projection spline adaptive filtering (MAPSAF) algorithm, which utilizes a modified Huber function with robustness against impulsive interference, and employs historical regression data to update the weight and the knot vector estimates for nonlinear filtering tasks. The detailed convergence and steady-state analyses of MAPSAF are also carried out in the mean and mean-square senses. In addition, an improved MAPSAF by exploiting the combined step sizes, called the CSS-MAPSAF algorithm, is derived to speed up the convergence on the premise of low steady-state misalignment. Numerical experiments in nonlinear system identification and nonlinear acoustic echo cancellation problems corroborate the theoretical performance analysis and the superiority of the proposed algorithms.
Conference Paper
This paper presents graph kernel adaptive filters that model nonlinear input-output relationships of streaming graph signals. To this end, we propose centralized and distributed graph kernel recursive least-squares (GKRLS) algorithms utilizing the random Fourier features (RFF) map. Compared with solutions based on the traditional kernel trick, the proposed RFF approach presents two significant advantages. First, it sidesteps the need to maintain a high-dimensional dictionary, whose dimension increases with the number of graph nodes and time, which renders prohibitive computational and storage costs, especially when considering least-squares algorithms involving matrix inverses. Second, the distributed algorithm developed in this paper, referred to here as the graph diffusion kernel recursive least-squares (GDKRLS) algorithm, does not require centralized dictionary training, making it ideal for distributed learning in dynamic environments. To examine the performance of the proposed algorithms, we analyze the mean convergence of the GDKRLS algorithm and conduct numerical experiments. Results confirm the superiority of the proposed RFF-based GKRLS and GDKRLS over their LMS counterparts.
Article
A linear-in-the-parameters nonlinear filter consists of a functional expansion block, which expands the input signal to a higher dimensional space nonlinearly, followed by an adaptive weight network. The number of weights to be updated depends on the type and order of the functional expansion used. When applied to a nonlinear system identification task, as the degree of the nonlinearity of the system is usually not known a priori , linear-in-the-parameters nonlinear filters are required to update a large number of coefficients to effectively model the nonlinear system. However, all the weights of the nonlinear filter may not contribute significantly to the identified model. We show via simulation experiments that, the weight vector of a linear-in-the-parameters nonlinear filter usually exhibits a low-rank nature. To take advantage of this observation, this paper proposes a class of linear-in-the-parameters nonlinear filters based on the nearest Kronecker product decomposition. The performance of the proposed filters is superior in terms of convergence behaviour as well as tracking ability in comparison to their traditional linear-in-the-parameters nonlinear filter counterparts, when tested for nonlinear system identification. Furthermore, the proposed nearest Kronecker product decomposition-based linear-in-the-parameters nonlinear filters has been shown to provide improved noise mitigation capabilities in a nonlinear active noise control scenario.
Article
Spline nonlinear adaptive filters are well known for their ability to efficiently model nonlinear systems while having low computational complexity. However, the performance of traditional spline adaptive filter degrades in the presence of impulsive disturbances. For improving the performance of spline adaptive filters in impulsive noise scenarios, a logarithmic hyperbolic cosine spline adaptive filter is proposed in this paper. To improve the convergence rate while maintaining low steady-state error, a variable parameter approach for the logarithmic hyperbolic cosine spline adaptive filter is also proposed. In addition, a robust nonlinear active noise control approach is also designed based on the proposed logarithmic hyperbolic cosine spline adaptive filter. The computational complexity of the proposed algorithms is studied and two approximate versions of the logarithmic hyperbolic cosine spline adaptive filter are proposed to decrease computational cost without degrading performance. Bound on learning rates are also derived to ensure the stability of the proposed adaptive systems. The simulation studies conducted, demonstrate the enhanced performance achieved by the proposed algorithms.
Article
Robust adaptive signal processing algorithms based on a generalized maximum correntropy criterion (GMCC) suffers from high steady state misalignment. In an endeavour to achieve lower steady state misalignment, in this letter we propose a generalized hyperbolic secant function (GHSF) as a robust norm and derive the generalized hyperbolic secant adaptive filter (GHSAF). The new algorithm is seen to offer robust system identification performance over the conventional GMCC algorithm. To further improve the convergence performance under non-Gaussian noise environments, we propose the nearest Kronecker product decomposition based GMCC and GHSAF algorithms. Extensive simulation study show the improved convergence performance provided by the proposed algorithms for system identification.
Article
The recursive least-squares (RLS) adaptive filter is an appealing choice in many system identification problems. The main reason behind its popularity is its fast convergence rate. However, this algorithm is computationally very complex, which may make it useless for the identification of long length impulse responses, like in echo cancellation. Computationally efficient versions of the RLS algorithm, like those based on the dichotomous coordinate descent (DCD) iterations or QR decomposition techniques, reduce the complexity, but still have to face the challenges related to long length adaptive filters (e.g., convergence/tracking capabilities). In this paper, we focus on a different approach to improve the efficiency of the RLS algorithm. The basic idea is to exploit the impulse response decomposition based on the nearest Kronecker product and low-rank approximation. In other words, a high-dimension system identification problem is reformulated in terms of low-dimension problems, which are combined together. This approach was recently addressed in terms of the Wiener filter, showing appealing features for the identification of low-rank systems, like real-world echo paths. In this paper, besides the development of the RLS algorithm based on this approach, we also propose a variable regularized version of this algorithm (using the DCD method to reduce the complexity), with improved robustness to double-talk. Simulations are performed in the context of echo cancellation and the results indicate the good performance of these algorithms.
Article
A new nonlinear filter, which employs an adaptive spline function as the basis function is designed in this paper. The input signal to this filter is used to generate suitable parameters to update the control points in a spline function. The update rule for updating the control points have been derived and a mean square analysis has been carried out. The output of the spline functions are suitably combined together to obtain the filter response. This filter is called the generalized spline nonlinear adaptive filter (GSNAF). The proposed GSNAF is similar to a functional link artificial neural network (FLANN), considering a functional expansion using spline basis functions. GSNAF has been shown to offer improved accuracy in benchmark classification scenarios and provide enhanced modeling accuracy in single input single output as well as in multiple input multiple output dynamic system identification cases.
Article
Although the LMS algorithm is often preferred in practice due to its numerous positive implementation properties, once the parameter space to estimate becomes large, the algorithm suffers of slow learning. Many ideas have been proposed to introduce some a-priori knowledge into the algorithm to speed up its learning rate. Recently also sparsity concepts have become of interest for such algorithms. In this contribution we follow a different path by focusing on the separability of linear operators, a typical property of interest when dealing with tensors. Once such separability property is given, a gradient type algorithm can be derived with significant increase in learning rate. Even if separability is only given to a certain extent, we show that the algorithm can still provide gains. We derive quality and quantity measures to describe the algorithmic behavior in such contexts and evaluate its properties by Monte Carlo simulations.
Article
Implementation of a feed-forward active noise control (ANC) system in a short duct may cause acoustic feedback between the active loudspeaker and the reference microphone. The conventional filtered-x least mean square (FxLMS) algorithm based ANC systems are not designed to handle this situation. Similarly, an FxLMS algorithm based ANC system fails to effectively mitigate noise when non-linearities are present in the system. In an endeavor to overcome these two limitations of traditional ANC systems, in this paper, we propose an adaptive infinite impulse response (IIR) spline filter for non-linear ANC. The adaptive nature of the spline activation function in the proposed scheme enables the filter to effectively switch from a linear to non-linear nature and vice versa, depending upon the scenario in which the ANC system is implemented.
Article
In this paper a novel class of nonlinear Hammerstein adaptive filters, consisting of a flexible memory-less function followed by a linear combiner, is presented. The nonlinear function involved in the adaptation process is based on a uniform cubic spline function that can be properly modified during learning. The spline control points are adaptively changed by using gradient-based techniques. This new kind of adaptive function is then applied to the input of a linear adaptive filter and it is used for the identification of Hammerstein-type nonlinear systems. In addition, we derive a simple form of the adaptation algorithm, an upper bound on the choice of the step-size and a lower bound on the excess mean square error in a theoretical manner. Some experimental results are also presented to demonstrate the effectiveness of the proposed method in the identification of high-order nonlinear systems.
Article
This paper introduces a new class of nonlinear adaptive filters, whose structure is based on Hammerstein model. Such filters derive from the functional link adaptive filter (FLAF) model, defined by a nonlinear input expansion, which enhances the representation of the input signal through a projection in a higher dimensional space, and a subsequent adaptive filtering. In particular, two robust FLAF-based architectures are proposed and designed ad hoc to tackle nonlinearities in acoustic echo cancellation (AEC). The simplest architecture is the split FLAF, which separates the adaptation of linear and nonlinear elements using two different adaptive filters in parallel. In this way, the architecture can accomplish distinctly at best the linear and the nonlinear modeling. Moreover, in order to give robustness against different degrees of nonlinearity, a collaborative FLAF is proposed based on the adaptive combination of filters. Such architecture allows to achieve the best performance regardless of the nonlinearity degree in the echo path. Experimental results show the effectiveness of the proposed FLAF-based architectures in nonlinear AEC scenarios, thus resulting an important solution to the modeling of nonlinear acoustic channels.
Article
In this paper a new class of nonlinear adaptive filters, consisting of a linear combiner followed by a flexible memory-less function, is presented. The nonlinear function involved in the adaptation process is based on a spline function that can be modified during learning. The spline control points are adaptively changed using gradient-based techniques. B-splines and Catmull-Rom splines are used, because they allow to impose simple constraints on control parameters. This new kind of adaptive function is then applied to the output of a linear adaptive filter and it is used for the identification of Wiener-type nonlinear systems. In addition, we derive a simple form of the adaptation algorithm and an upper bound on the choice of the step-size. Some experimental results are also presented to demonstrate the effectiveness of the proposed method.