Gradient Adaptive Filter Research Articles

State-of-charge estimation with adaptive extended Kalman filter and extended stochastic gradient algorithm for lithium-ion batteries

Online state-of-charge (SOC) estimation is a critical element for battery management systems and it requires lower computing cost and acceptable range of accuracy. This paper proposes a new model-based SOC estimation method for lithium-ion batteries. By utilizing the state estimation to identify the model parameters and then re-estimate the state by using the identified parameters, the two steps of parameter identification and state estimation are integrated into one closed-loop algorithm and they are implemented by using extended stochastic gradient (ESG) algorithm and adaptive extended Kalman filter (AEKF), respectively. In this method, it is unnecessary to calculate each circuit parameter of the model separately resulting in simper structure and lower computing cost. Experimental results indicate that the proposed SOC estimation algorithm has good performance in terms of estimation accuracy and robustness under different test conditions. It is therefore more suitable for online SOC estimation of lithium-ion batteries.

Anchor-Based Segmentation of Periventricular White Matter for Neonatal HIE

The segmentation and quantitative features of periventricular white matter injury in magnetic resonance imaging are helpful for determining an accurate diagnosis of neonatal hypoxia ischemia encephalopathy (HIE) in clinical practice, but they are tedious tasks. In this paper, we present an interactive segmentation method for clinical magnetic resonance imaging (MRI) using sample estimation at the subvoxel level to establish a quantitative diagnosis. The coarse-grained voxels in clinical MRI are divided into subvoxels, and the intensities of subvoxels are adjusted using gradient adaptive filtering. Because the cerebrospinal fluid is more obvious in grayscale than the white matter and gray matter on T1-weighted images (T1WI), T2-weighted images (T2WI) and fluid-attenuated inversion recovery (FLAIR) images, and its population variance follows a Chi-square distribution, we segment the periventricular region as an anchor by estimating the parameters of the selected data and expand the anchor to obtain periventricular white matter of various widths. After aligning the segmented periventricular white matter in various modalities, we compute the radiomics features of regions of interest for the quantitative diagnosis of HIE. Current clinical practice confirms that our approach is effective and satisfies the clinical diagnostic requirements for neonatal HIE.

Conjugate gradient adaptive matched filter

We consider an adaptive reduced-rank detector, referred to as the CG-AMF detector, which is obtained by using the conjugate gradient (CG) algorithm to solve for the weight vector of the adaptive matched filter (AMF). The CG is a computationally efficient iterative algorithm, which finds the projection of the AMF weight vector to the Krylov subspace with a dimension growing with the CG iterations. This effectively leads to a family of reduced-rank detectors indexed by the number of CG iterations. The main purpose of this paper is to examine the output signal-to-interference-and-noise ratio (SINR) of the CG-AMF detector in the presence of strong clutter/interference. Specifically, by exploiting a connection between the CG algorithm and the Lanczos algorithm, we show the output SINR can be asymptotically expressed in a simple form involving a Ritz vector of the sample covariance matrix. The probability density function (pdf) and expected value of the output SINR are then obtained based on this approximation. Our theoretical analysis of the CG-AMF detector is verified by computer simulation. Numerical comparisons are also made with several popular reduced-rank detectors using either data-independent or data-dependent rank reduction approaches. Our results show that for a fixed training size, the CG-AMF detector often reaches its peak output SINR with a lower rank compared with the other reduced-rank detectors, which implies that the CG-AMF detector has lower computational complexity and less training requirement.

Self-whitening algorithms for adaptive equalization and deconvolution

In equalization and deconvolution tasks, the correlated nature of the input signal slows the convergence speeds of stochastic gradient adaptive filters. Prewhitening techniques have been proposed to improve the convergence performance, but the additional coefficient memory and updates for the prewhitening filter can be prohibitive in some applications. We present two simple algorithms that employ the equalizer as a prewhitening filter within the gradient updates. These self-whitening algorithms provide quasi-Newton convergence locally about the optimum coefficient solution for deconvolution and equalization tasks. Multichannel extensions of the techniques are also described.

Improved stochastic gradient adaptivefilter with gradient adaptive step size

The performance of a stochastic gradient adaptive filter can be significantly improved by introducing a forgetting factor. The complexity of the original algorithm can also be reduced by using only the signs of error signals and input signals in the gradient adaptive step size computation.

A robust mixed-norm adaptive filter algorithm

We propose a new member of the family of mixed-norm stochastic gradient adaptive filter algorithms for system identification applications based upon a convex function of the error norms that underlie the least mean square (LMS) and least absolute difference (LAD) algorithms. A scalar parameter controls the mixture and relates, approximately, to the probability that the instantaneous desired response of the adaptive filter does not contain significant impulsive noise. The parameter is calculated with the complementary error function and a robust estimate of the standard deviation of the desired response. The performance of the proposed algorithm is demonstrated in a system identification simulation with impulsive and Gaussian measurement noise.

A unifying treatment of general gradient adaptive digital filter and convergence analysis with arbitrary inputs

AbstractThe unifying treatment of a general gradient adaptive algorithm studied here is based on the principles of minimum mean square method and instrumental variable method by formulating the general oblique projection algorithm, including finite impulse response (FIR) and serial‐parallel infinite impulse response (IIR)‐type configurations. the oblique projection algorithm of low‐order past‐value data space is derived in which the dimension of past‐value data space can be easily adjusted, including the least mean square (LMS) method, learning‐identification method and gradient instrumental variable method (GIVE). Next, the new analysis model which describes the convergence process of general gradient algorithm is shown and the convergence conditions are indicated for the first and second moments with arbitrary inputs. It is found that this condition is more severe as compared with the white input. Moreover, the process of convergence of these conditions is studied by simulation.

LMS+F algorithm

A stochastic gradient adaptive filter algorithm using a time-varying mixed criterion is proposed. The algorithm optimises a cost function based on an adaptively adjusted combination of the LMF and LMS functions. Two different combination strategies are studied and sufficient convergence conditions are presented. The time-varying nature of the cost function adds flexibility and enables rapid response of the algorithm.

Convergence analysis of a data echo canceller with a stochastic gradient adaptive FIR filter using the sign algorithm

Under the assumption that the "errors at the taps" are Gaussian distributed, a new set of recurrence formulae is derived for calculating theoretical convergence process of a data echo canceller equipped with an FIR filter that is adaptively controlled by using the "stochastic gradient sign algorithm" with a binary and white process as the filter input. Convergence curves for the mean squared residual echo based on the recurrence formulae show an excellent agreement with those obtained by simulation. Approximate recurrence formulae that yield a useful, though less accurate, estimation of the residual echo convergence are also proposed. Furthermore, a closed-form solution to the approximate recurrence formulae is derived. And finally, probability density function of the residual echo is depicted as it changes its shape as the echo canceller converges.

Least mean mixed-norm adaptive filtering

A new family of stochastic gradient adaptive filter algorithms is proposed which is based on mixed error norms. These algorithms combine the advantages of different error norms, for example the conventional, relatively well-behaved, least mean square algorithm and the more sensitive, but better converging, least mean fourth algorithm. A mixing parameter is included which controls the proportions of the error norms and offers an extra degree of freedom within the adaptation. A system identification simulation is used to demonstrate the performance of a least mean mixed-norm (square and fourth) algorithm.

Fast multidelay block transform-domain adaptive filters based on a two-dimensional optimum block algorithm

Three fast and flexible block transform-domain gradient adaptive digital filters (ADF's) for finite impulse response (FIR) adaptive filtering are presented. The proposed ADF's employ a two-dimensional optimum block algorithm (OBA) in order to obtain a time-varying convergence factor which is optimized in a least square (LS) sense for fast and accurate adaptation, and to allow one to choose an appropriate transform size for smaller block delay and more efficient use of a hardware. In the first one of the proposed ADF's, the multidelay frequency-domain OBA (MFOBA), the two-dimensional OBA is realized with the conventional fast Fourier transform (FFT). In the second one, the normalized MFOBA (NMFOBA), a robust normalization method significantly improves its convergence speed for colored input signals. In the third one, the Fermat number transform (FNT) based realization (MFOBA/FNT), the desirable properties of the FNT (e.g., no roundoff errors, no multiplication and no complex basis functions) are utilized to yield an ideal structure for an efficient hardware implementation. By computer simulation, it is shown that these ADF's are faster than other transform-domain adaptive algorithms.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Adaptive polynomial filters

Adaptive nonlinear filters equipped with polynomial models of nonlinearity are explained. The polynomial systems considered are those nonlinear systems whose output signals can be related to the input signals through a truncated Volterra series expansion or a recursive nonlinear difference equation. The Volterra series expansion can model a large class of nonlinear systems and is attractive in adaptive filtering applications because the expansion is a linear combination of nonlinear functions of the input signal. The basic ideas behind the development of gradient and recursive least-squares adaptive Volterra filters are first discussed. Adaptive algorithms using system models involving recursive nonlinear difference equations are then treated. Such systems may be able to approximate many nonlinear systems with great parsimony in the use of coefficients. Also discussed are current research trends and new results and problem areas associated with these nonlinear filters. A lattice structure for polynomial models is described.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Performance analysis of adaptive filters equipped with the dual sign algorithm

A convergence analysis for stochastic gradient adaptive filters equipped with the dual sign algorithm (DSA) is presented. Expressions for the mean and mean-squared values of the coefficient misalignment vector are derived under the assumption that the input signal is Gaussian. The main differences between the current analysis and previous analyses are: (1) the present analysis is valid for arbitrary autocorrelation matrices, and (2) the analysis does not approximate the behavior for the DSA as switching between two sign algorithms a certain amount of time after the DSA is initialized. A simulation example comparing the analytical results with empirical ones is presented and the two curves show excellent agreement with each other.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Fixed-point error analysis of stochastic gradient adaptive lattice filters

A theoretical analysis is presented of the stochastic gradient adaptive lattice filter used as a linear, one-step predictor, when the effects of finite-precision arithmetic are taken into account. Only the fixed-point implementation is considered. Both the unnormalized and normalized adaptation algorithms are analyzed. Expressions of the steady-state mean-squared values of the accumulated numerical errors in the computation of the reflection coefficients and the prediction errors of different orders have been developed. The results show that the dominant term in the expressions for the mean-squared values of the numerical errors is inversely proportional to the convergence parameter. Furthermore, they indicate that the quantization errors associated with the reflection coefficients are more critical than those associated with representing the prediction-error sequences. Signals with high correlation among samples produce larger numerical errors in the adaptive lattice filter. The authors present several simulation examples that show close agreement with the theoretical results. They also show that the gradient adaptive lattice filters have better numerical properties than their transversal counterparts. >

An adaptive algorithm for short-term multinode load forecasting in power systems

An online adaptive escalator lattice structure for orthogonalization of multiple-channel signals is used to predict load demands among loading nodes in a power system by an autoregressive multiple-channel mode. Since the escalator outputs are white and also uncorrelated with each channel or node, the parameters of the algorithm are updated adaptively using scalar operations. Because matrix or vector operations are not required in the updating procedures, the convergence speed is insensitive to the ratio of the largest to the smallest eigenvalues of the loads' covariance matrix. Thus the prediction filter has a faster convergence rate than common matrix-oriented gradient adaptive filters. Computer simulation shows that this algorithm has a faster convergence rate and better numerical properties in the adaptive process. This is very attractive for multinode load forecasting in a large power system, where each load model varies with time and has different statistical characteristics, or where loads are nonstationary and the ratio of eigenvalues in the load covariance matrix is large. >

Improved convergence analysis of stochastic gradient adaptive filters using the sign algorithm

Convergence analysis of stochastic gradient adaptive filters using the sign algorithm is presented in this paper. The methods of analysis currently available in literature assume that the input signals to the filter are white. This restriction is removed for Gaussian signals in our analysis. Expressions for the second moment of the coefficient vector and the steady-state error power are also derived. Simulation results are presented, and the theoretical and empirical curves show a very good match.

Application of gradient adaptive lattice filters to channel equalisation

In this paper, the potential application of adaptive lattice equalisers based on gradient search adaptive algorithms, is examined. The assessment is based on simulations of equaliser performance in data communications systems which provides a comparison of the relative performance of these adaptive lattice equalisers with transversal equalisers. It highlights the critical balance between rapid convergence and degradation due to algorithm noise, which is involved when selecting the algorithm stepsize. Two new adaptive equaliser approaches are suggested, one based on a timed lattice structure feeding a linear combiner. The other is a short timed lattice structure in cascade with a transversal equaliser. Both approaches are shown to offer fast converging adaptive equalisers for data communications applications, which have a converged error performance superior to that of the direct application of a gradient adaptive lattice equaliser.