A signal modeling approach is proposed to reduce the problem of estimating frequencies of noisy sinusoids to the parameter estimation problem. Both white and colored noise are considered. The maximum likelihood approa...
详细信息
A signal modeling approach is proposed to reduce the problem of estimating frequencies of noisy sinusoids to the parameter estimation problem. Both white and colored noise are considered. The maximum likelihood approach leads to an algorithm computationally inefficient for low values of signal-to-noise ratio. Based on the equation-error formulation, an iterative inverse filtering algorithm is derived in the case of colored noise, and a generalized least-squares algorithm in the case of white noise. It is shown, on the basis of numerous experimental results, that the iterative inverse filtering algorithm provides highly accurate estimates of unknown frequencies for low values of signal-to-noise ratio, even in the case of a small number of sampling points.
Aiming at exploiting temporal correlations across consecutive time frames in the short-time Fourier transform (STFT) domain, multi-frame algorithms for single-microphone speech enhancement have been proposed. Typicall...
详细信息
Aiming at exploiting temporal correlations across consecutive time frames in the short-time Fourier transform (STFT) domain, multi-frame algorithms for single-microphone speech enhancement have been proposed. Typically, the multi-frame filter coefficients are either estimated directly using deep neural networks or a certain filter structure is imposed, e.g., the multi-frame minimum variance distortionless response (MFMVDR) filter structure. Recently, it was shown that integrating the fully differentiable MFMVDR filter into an end-to-end supervised learning framework employing temporal convolutional networks (TCNs) allows for a high estimation accuracy of the required parameters, i.e., the speech inter-frame correlation vector and the interference covariance matrix. In this paper, we investigate different covariance matrix structures, namely Hermitian positive-definite, Hermitian positive-definite Toeplitz, and rank-1. The main differences between the considered matrix structures lie in the number of parameters that need to be estimated by the TCNs as well as the required linear algebra operations. For example, assuming a rank-1 matrix structure, we show that the MFMVDR filter can be written as a linear combination of the TCN outputs, significantly reducing computational complexity. In addition, we consider a covariance matrix estimation procedure based on recursive smoothing. Experimental results on the deep noise suppression challenge dataset show that the estimation procedure using the Hermitian positive-definite matrix structure yields the best performance, closely followed by the rank-1 matrix structure at a much lower complexity. Furthermore, imposing the MFMVDR filter structure instead of directly estimating the multi-frame filter coefficients slightly but consistently improves the speech enhancement performance.
This paper derives a theory for adaptive filters which operate on filter bank outputs, here called filter bank adaptive filters (FBAF's). It is shown how the FBAF's are a generalization of transform domain ada...
详细信息
This paper derives a theory for adaptive filters which operate on filter bank outputs, here called filter bank adaptive filters (FBAF's). It is shown how the FBAF's are a generalization of transform domain adaptive filters and adaptive filters based on structural subband decompositions. The minimum mean-square error performance and convergence properties of FBAF's are determined as a function of filter bank used. A parametrization for a class of FIR perfect reconstruction filter banks is derived which is used to design FBAF's having optimal error performance given prior knowledge of the application. Simulations are performed to illustrate the derived theory and demonstrate the improved error performance of the FBAF's relative to the LMS algorithm, when prior knowledge is incorporated.
An important consideration in the design of an adaptive system is the convergence rate of the system. This is closely related to the system's ability to track a time-varying optimum. Basic adaptive filtering algor...
详细信息
An important consideration in the design of an adaptive system is the convergence rate of the system. This is closely related to the system's ability to track a time-varying optimum. Basic adaptive filtering algorithms give poor convergence performance when the input to the adaptive system is colored. There are more sophisticated algorithms which converge very rapidly regardless of the input spectrum, but these algorithms typically require O(N2) computation, where N is the order of the adaptive filter. This is a significant disadvantage for real-time applications, especially where economic constraints must be met. Also, many of these algorithms have a reputation for behaving poorly in finite precision implementation. In this paper, an adaptive filtering algorithm is introduced which employs a quasi-Newton approach to give rapid convergence even with colored inputs. The algorithm achieves an overall computational requirement of O(N). And this fast quasi-Newton (FQN) algorithm appears to be quite robust in finite precision implementations.
A chip set for 2D subband filtering of HDTV signals has been designed, fabricated and successfully tested. The two chips perform 10 * 14 quadrature mirror filtering for analysis filtering at the coder and synthesis fi...
详细信息
A chip set for 2D subband filtering of HDTV signals has been designed, fabricated and successfully tested. The two chips perform 10 * 14 quadrature mirror filtering for analysis filtering at the coder and synthesis filtering at the decoder. In order to achieve a very compact realization, the architectures utilize all a priori known properties of the filter algorithm. A 2D polyphase filter structure reduces the processing clock rate from the 72-MHz sampling rate to a moderate 18 MHz. The memory for vertical filtering is realized by on-chip parallel shift registers with multiphase clocking. A small silicon area for the filter arithmetic is achieved by application of carry save adder trees with fixed filter coefficients represented by canonical signed digits. A complete filterbank for luminance and chrominance signals consists of four identical chips, each with 450000 transistors on 92 mm2.
A new transform domain adaptive digital filter using the common LMS adaptation algorithm based on a highly efficient circular convolution algorithm is presented. The proposed filter has an easily implementable structu...
详细信息
A new transform domain adaptive digital filter using the common LMS adaptation algorithm based on a highly efficient circular convolution algorithm is presented. The proposed filter has an easily implementable structure involving only real-valued arithmetic. It is shown to converge to the Wiener solution in the transform domain and has significant advantage in efficiency over the corresponding FFT-based adaptive digital filters.
This article provides an overview of nonlinear statistical signal processing based on the Bayesian paradigm. The next-generation processors are well founded on MC simulation-based sampling techniques. The development ...
详细信息
This article provides an overview of nonlinear statistical signal processing based on the Bayesian paradigm. The next-generation processors are well founded on MC simulation-based sampling techniques. The development of the sequential Bayesian processor is reviewed using the state-space models. The popular bootstrap algorithm was outlined and applied to an ocean acoustic synthetic aperture towed array target tracking problem to test the performance of a particle filtering technique.
Fan-beam collimators are used in single photon emission computed tomography to improve the sensitivity for imaging of small organs. The disadvantage of fan-beam collimation is the truncation of projection data surroun...
详细信息
Fan-beam collimators are used in single photon emission computed tomography to improve the sensitivity for imaging of small organs. The disadvantage of fan-beam collimation is the truncation of projection data surrounding the organ of interest or, in those cases of imaging large patients, of the organ itself producing reconstruction artifacts. A spatially varying focal length fan-beam collimator has been proposed to eliminate the truncation problem and to maintain good sensitivity for the organ of interest. The collimator is constructed so that the focal lengths of the holes vary across the face of the collimator with the shortest focal lengths at the center and the longer focal lengths at the periphery of the collimator. The variation of the focal length can have various functional forms but in our work it is assumed to increase monotonically toward the edge of the collimator. A backprojection filtering reconstruction algorithm is derived for this type of collimation. The algorithm first backprojects the projections, then performs a two-dimensional filtering. The algorithm is efficient and has been tested via computer simulations.
Recent advances in the utilization of Lie Groups for robotic localization have led to dramatic increases in the accuracy of estimation and uncertainty characterization. One of the novel methods, the Invariant Extended...
详细信息
Recent advances in the utilization of Lie Groups for robotic localization have led to dramatic increases in the accuracy of estimation and uncertainty characterization. One of the novel methods, the Invariant Extended Kalman Filter (InEKF) extends the Extended Kalman Filter (EKF) by leveraging the fact that some error dynamics defined on matrix Lie Groups satisfy a log-linear differential equation. Utilization of these observations result in linearization with minimal approximation error, no dependence on current state estimates, and excellent convergence and accuracy properties. In this letter we show that the primary sensors used for underwater localization, inertial measurement units (IMUs) and doppler velocity logs (DVLs) meet the requirements of the InEKF. Furthermore, we show that singleton measurements, such as depth, can also be used in the InEKF update with minor modifications, thus expanding the set of measurements usable in an InEKF. We compare convergence, accuracy and timing results of the InEKF to a quaternion-based EKF using a Monte Carlo simulation and show notable improvements in long-term localization and much faster convergence with negligible difference in computation time.
The problem of tracking time-varying parameters of a linear stochastic system is considered, and an identification method based on parameter estimation and filtering is described. The proposed algorithm combines the s...
详细信息
The problem of tracking time-varying parameters of a linear stochastic system is considered, and an identification method based on parameter estimation and filtering is described. The proposed algorithm combines the standard weighted least squares (WLS) identification with low-pass filtering of parameter estimates. It is shown that the parameter tracking properties of the combined estimation-filtering method are exactly the same as the tracking capabilities of the WLS estimator characterized by the appropriately defined weighting sequence and can be analyzed in terms of the associated frequency characteristics. The main advantage of the method is that it allows for efficient implementation of banks of adaptive filters characterized by different memory lengths without compromising the good tracking capabilities of WLS estimators. Additionally, it provides the designer with much greater flexibility in shaping the window.< >
暂无评论