Multichannel Least Mean Squares Research Articles

Blind Deconvolution of Ultrasound Images Using l1 -Norm-Constrained Block-Based Damped Variable Step-Size Multichannel LMS Algorithm.

The problem of improving the ultrasound image resolution by undoing the effect of convolution on backscattered radio-frequency (RF) data caused by the point spread function (PSF) of ultrasonic imaging system is one of the key problems in the reconstruction of the medical ultrasound images. In this paper, the tissue reflectivity functions (TRFs) are directly estimated from the noisy and nonstationary RF data using the block-based multichannel least-mean square ( l1 -bMCLMS) algorithm without any prior knowledge of the PSF. To account for the nonstationarity and incomplete acquisition problem of the ultrasound RF data a modified block-based cross-relation equation has been developed. An l1 -norm regularized cost function based on the proposed modified cross-relation equation is then formulated for blind estimation of the TRFs using the new l1 -bMCLMS algorithm. A damped variable step-size is also developed to compensate for the noise effect and to improve the convergence speed of the algorithm. The PSF is then estimated from multiple lateral blocks of RF data using the regularized multiple-input/output inverse theorem, which is known to be suitable for both minimum and nonminimum phase signals. The salient feature of the proposed method is that no basis function is required for TRFs and/or PSF. The efficacy of the proposed method is examined using the simulation/experimental phantom data and in vivo RF data and evaluated in terms of the quality metrics: resolution gain (RG), normalized projection misalignment (NPM), and shifted normalized mean square error (snMSE). The results show that the RG and NPM improvements of TRFs estimation of 0.12∼5.2 and 3.34∼22.82 dB, respectively, and the snMSE improvement of the PSF estimation of the order 10(2∼4) can be achieved in our technique as compared with the other techniques in the literature.

Deep neural network-based bottleneck feature and denoising autoencoder-based dereverberation for distant-talking speaker identification

AbstractDeep neural network (DNN)-based approaches have been shown to be effective in many automatic speech recognition systems. However, few works have focused on DNNs for distant-talking speaker recognition. In this study, a bottleneck feature derived from a DNN and a cepstral domain denoising autoencoder (DAE)-based dereverberation are presented for distant-talking speaker identification, and a combination of these two approaches is proposed. For the DNN-based bottleneck feature, we noted that DNNs can transform the reverberant speech feature to a new feature space with greater discriminative classification ability for distant-talking speaker recognition. Conversely, cepstral domain DAE-based dereverberation tries to suppress the reverberation by mapping the cepstrum of reverberant speech to that of clean speech with the expectation of improving the performance of distant-talking speaker recognition. Since the DNN-based discriminant bottleneck feature and DAE-based dereverberation have a strong complementary nature, the combination of these two methods is expected to be very effective for distant-talking speaker identification. A speaker identification experiment was performed on a distant-talking speech set, with reverberant environments differing from the training environments. In suppressing late reverberation, our method outperformed some state-of-the-art dereverberation approaches such as the multichannel least mean squares (MCLMS). Compared with the MCLMS, we obtained a reduction in relative error rates of 21.4% for the bottleneck feature and 47.0% for the autoencoder feature. Moreover, the combination of likelihoods of the DNN-based bottleneck feature and DAE-based dereverberation further improved the performance.

Distant-talking speaker identification by generalized spectral subtraction-based dereverberation and its efficient computation

Abstract Previously, a dereverberation method based on generalized spectral subtraction (GSS) using multi-channel least mean-squares (MCLMS) has been proposed. The results of speech recognition experiments showed that this method achieved a significant improvement over conventional methods. In this paper, we apply this method to distant-talking (far-field) speaker recognition. However, for far-field speech, the GSS-based dereverberation method using clean speech models degrades the speaker recognition performance. This may be because GSS-based dereverberation causes some distortion between clean speech and dereverberant speech. In this paper, we address this problem by training speaker models using dereverberant speech obtained by suppressing reverberation from arbitrary artificial reverberant speech. Furthermore, we propose an efficient computational method for a combination of the likelihood of dereverberant speech using multiple compensation parameter sets. This addresses the problem of determining optimal compensation parameters for GSS. We report the results of a speaker recognition experiment performed on large-scale far-field speech with different reverberant environments to the training environments. The proposed GSS-based dereverberation method achieves a recognition rate of 92.2%, which compares well with conventional cepstral mean normalization with delay-and-sum beamforming using a clean speech model (49.0%) and a reverberant speech model (88.4%). We also compare the proposed method with another dereverberation technique, multi-step linear prediction-based spectral subtraction (MSLP-GSS). The proposed method achieves a better recognition rate than the 90.6% of MSLP-GSS. The use of multiple compensation parameters further improves the speech recognition performance, giving our approach a recognition rate of 93.6%. We implement this method in a real environment using the optimal compensation parameters estimated from an artificial environment. The results show a recognition rate of 87.8% compared with 72.5% for delay-and-sum beamforming using a reverberant speech model.

Algorithmic Trading Using Phase Synchronization

A novel trading algorithm which performs trading decisions by making use of phase synchronization between oscillatory components of asset pairs is proposed. This way the phase information ascertains a leading asset, which is then used to predict the lagging asset. The oscillatory components of asset pairs are identified using the Synchrosqueezed Transform (SST), which facilitates stable and online implementation. The performance of the proposed approach is compared with existing algorithms used extensively by traders, such as the moving average cross-over, and an extrapolation algorithm based upon the multichannel-least mean square (MLMS).

Cross-Relation-Based Blind SIMO Identifiability in the Presence of Near-Common Zeros and Noise

The blind identification of single-input multiple-output (SIMO) systems is often performed by exploiting a cross-relation (CR) between channel pairs. It has been shown that this property allows an estimation of channel impulse responses up to a common gain factor if certain identifiability conditions are met. In this case, the estimated channels can be evaluated by a gain-compensated system distance known as normalized projection misalignment (NPM). Current algorithms for blind channel identification, however, suffer in the presence of insufficient channel diversity and observation noise. In this paper, we first demonstrate that in the absence of noise the CR identification error for channels with exact common zeros is given by a single-channel pole-zero transfer function. Next, we extend our analysis to the realistic case of near-common zeros and noise for which we show that the effective error can still be approximated by a common transfer function as long as the distance between the channel zeros remains below a signal-to-noise ratio-dependent threshold. A finite impulse response (FIR) modeling of the error then enables us to define a common-filter-error-compensated system distance, termed normalized filter-projection misalignment (NFPM), which establishes a natural extension to the NPM analysis. By finally considering realistic channels, which we blindly estimate with the adaptive multichannel least mean-square (MCLMS) algorithm, we demonstrate that the NFPM reliably reaches the noise floor, confirming that the effective error can be approximated by a common FIR filter.

Distant-Talking Speech Recognition Based on Spectral Subtraction by Multi-Channel LMS Algorithm

SUMMARY We propose a blind dereverberation method based on spectral subtraction using a multi-channel least mean squares (MCLMS) algorithm for distant-talking speech recognition. In a distant-talking environment, the channel impulse response is longer than the short-term spectral analysis window. By treating the late reverberation as additive noise, a noise reduction technique based on spectral subtraction was proposed to estimate the power spectrum of the clean speech using power spectra of the distorted speech and the unknown impulse responses. To estimate the power spectra of the impulse responses, a variable step-size unconstrained MCLMS (VSS-UMCLMS) algorithm for identifying the impulse responses in a time domain is extended to a frequency domain. To reduce the effect of the estimation error of the channel impulse response, we normalize the early reverberation by cepstral mean normalization (CMN) instead of spectral subtraction using the estimated impulse response. Furthermore, our proposed method is combined with conventional delay-andsum beamforming. We conducted recognition experiments on a distorted speech signal simulated by convolving multi-channel impulse responses with clean speech. The proposed method achieved a relative error reduction rate of 22.4% in relation to conventional CMN. By combining the proposed method with beamforming, a relative error reduction rate of 24.5% in relation to the conventional CMN with beamforming was achieved using only an isolated word (with duration of about 0.6s) to estimate the spectrum of the impulse response.

Robust multichannel least mean square-type algorithms with fast decaying transient for blind identification of acoustic channels

The multichannel least mean square (MCLMS) is an attractive and effective algorithm for blind channel identification in the noise-free case. Some recent studies show that the performance of the MCLMS algorithm significantly deteriorates in a noisy environment, that is, the blind MCLMS solution does not remain collinear with the channel vector. Therefore the authors propose non-conventional technique that helps the MCLMS algorithm converge to a novel steady-state solution that is a weighted combination of all the eigenvectors, with the weight profile inversely proportional to the eigenvalues. The improved performance of the proposed solution is verified both analytically and numerically. The algorithm is then optimised by introducing an adaptive step size that ensures fast decay of the transient response, giving stability as well as rapid convergence to the final solution. The authors then apply the proposed technique to different variants of the MCLMS algorithm, including frequency-domain implementations, to achieve a noise-robust performance. Computer simulations are presented that show improved performance of the proposed algorithms for blind identification of both acoustic and random channels with noise.

Generalized Optimal Step-Size for Blind Multichannel LMS System Identification

The choice of step-size in adaptive blind channel identification using the multichannel least mean squares (MCLMS) algorithm is critical and controls its convergence rate, stability, and sensitivity to noise. In this letter, we derive the expression for an optimal step-size in the Wiener sense and investigate its properties. An implementation technique for the Wiener solution of the self-adaptive step-size is presented, and it is shown that significant performance improvements are obtained compared to existing approaches in the presence of noise

Optimal step size of the adaptive multichannel LMS algorithm for blind SIMO identification

Adaptive algorithms for blindly identifying single-input multiple-output (SIMO) systems are appealing because of their computational efficiency and capability of continuously tracking a time-varying system. Adaptive multichannel least-mean-square (MCLMS) algorithms (with and without the unit-norm constraint) are analyzed, and the optimal step size is derived. A simple yet effective variable step-size MCLMS algorithm is proposed, and its performance is evaluated with simulations.

A relaxed condition for "perfect" cancellation of broadband noise in 3D enclosures

This paper presents a relaxed condition for "perfect" cancellation of broadband noise in 3D enclosures. On the basis of a truncated modal model, it can be shown that the primary and secondary paths belong to a same subspace if a certain condition is satisfied. There exists a finite impulse response (FIR) filter transfer function vector for perfect cancellation of the primary paths. The analytical result is verified numerically with an active noise control (ANC) system in a 3D rectangular enclosure. The proposed ANC scheme is shown to fit well into the framework of an existing multichannel least-mean squares (LMS) algorithm for adaptive implementation.

Multichannel RLS algorithms and FTF algorithms for active noise control and sound reproduction systems

In the fields of active noise control (ANC) and transaural sound reproduction (TSR), multichannel FIR adaptive filters are extensively used. For the learning of such FIR adaptive filters, recursive-least-squares (RLS) algorithms are known to typically produce a faster convergence speed than stochastic gradient descent techniques, such as the basic least-mean-squares (LMS) algorithm or even the fast convergence Newton-LMS, gradient-adaptive-lattice (GAL) LMS and discrete-cosine-transform (DCT) LMS algorithms. In this presentation, multichannel RLS algorithms and multichannel fast-transversal-filter (FTF) algorithms are introduced, with the structures of some stochastic gradient descent algorithms used in ANC: the filtered-x LMS, the adjoint-LMS and the modified filtered-x LMS. The new algorithms can be used in ANC systems or for the deconvolution of sounds in TSR systems. Also, heuristic techniques are introduced, to compensate for the potential ill-conditioning of the correlation matrix in ANC or TSR systems, and for the potential numerical instability of the multichannel FTF-based algorithms. Simulations of ANC and TSR systems will compare the performance of the different multichannel LMS, RLS, and FTF-based algorithms.