一种基于软判决的立体声回声抵消算法

提出了一种新的基于维纳滤波的频域算法来解决立体声回声抵消问题,该算法不需要对立体声信号预处理,从而最大程度地保证了近端语音质量。并且它具有很好的鲁棒性,很快的收敛速度和跟踪速度,因而具有一定的实用价值。引入了语音增强中的软判决方法来进一步提高算法的性能。新的算法在保证近端语音质量的同时达到了更好的回声压制效果。仿真实验证明了新算法的良好性能。     

基于多相滤波器组的非线性回声抵消

在免提通话系统和移动通信设备中，扬声器常常工作在较高的音量下，容易发生过载现象，从而产生明显的非线性声学回声，这在小微型扬声器中更加常见。常用的线性AEC(Acoustic Echo Cancellation)算法无法消除此类非线性回声，因此通话质量受到严重影响。非线性回声主要表现为额外的高频谐波分量，这些分量使得全带系统不再满足线性关系，而通常的AEC算法都是基于最小化全带误差推导而来，因此性能很容易受到非线性失真的影响。本文提出了一种基于多相滤波器组的子带AEC算法，把全带误差变成了各个子带的误差，因而把谐波失真成分变成了某些子带内的加性噪声，这使得谐波失真较小的那些子带依然能够正常收敛。通过仿真和实测实验，当出现非线性失真时，新方法的ERLE(Echo Return Loss Enhancement)明显高于经典的全带时域和频域方法，对于非线性失真明显的语音信号，ERLE提升约10 dB。      

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.     

基于无噪后验错误矢量信号能量的变正则因子仿射投影算法

变正则因子技术是提高仿射投影自适应算法性能的重要方法之一.由于环境噪声的影响,现有的变正则因子自适应算法收敛速度较慢且稳态误差较大,各种测量、评估误差的存在进一步恶化了算法性能.为提高自适应算法的跟踪性能,本文在分析无噪先验错误矢量、无噪后验错误矢量和额外均方错误间关系的基础上,提出通过最小化无噪后验错误矢量信号能量来推导自适应变正则因子表达式的方法.在实践应用中,该方法利用了测量噪声的统计方差特性,并提出一种更加光滑且更加容易控制的指数缩放因子评估方法.系统辨识的仿真结果表明本文方法与传统的变正则因子方法以及变步长方法相比有更快的收敛速度与更低的稳态误差.     

基于频域逐级回归的声学回声控制

传统声学回声控制算法一般采用基于随机梯度法更新的频域分块自适应滤波(PBFDAF)方法,但在以语音为主要回声信号的室内混响环境中,由于回声路径不稳定,往往收敛速度较慢,难以实现足够的回声抑制。该文提出一种基于频域逐级回归的声学回声控制算法。通过逐级回归分析远端信号和麦克风信号之间的线性关系,可以在保持较小的偏差的同时实现收敛较快的系统估计。同时,由于逐级分析了两通道间的短时相干性,因而该算法无需像常见方法一样,额外进行基于通道间相干函数的残余回声抑制或双讲检测,从而保持系统的紧凑性。若进一步假定近端背景噪声准平稳,则可利用基于近端信号非平稳程度的自适应平滑因子,在实现系统估计快速收敛的同时确保其稳定性。实验表明,该方法在常见的近端环境噪声水平下,在收敛速度和稳态误差上相对传统方法有显著优势,非常适合应用在室内远讲模式下的声学回声控制中。     

Signal subspace approach for narrowband noise reduction in speech

A signal subspace method is proposed for speech enhancement in the presence of narrowband noise. A fundamental assumption in subspace methods for noise reduction is that the noise covariance matrix is positive definite. However, this is not always the case, especially when the noise has narrowband characteristics. Based on the eigenvalue decomposition of the rank deficient noise covariance matrix, it is shown how to formulate the enhancement algorithm by decomposing the vector space of noisy signal into a signal-plus-noise subspace and a noise-free subspace. The proposed subspace partition is different from the conventional subspace approaches in that the noise reduction algorithm is implemented using the whitening approach exclusively in the signal-plus-noise subspace. The enhancement is performed by estimating the clean speech from the signal-plus-noise subspace and adding the components in the noise-free subspace. An explicit form of the estimator is presented, and examples are illustrated to validate the effectiveness of the proposed method.     

Choix d’un algorithme en précision finie,pour annuleur d’écho

The comparison of two different algorithms for adaptive echo cancellers is analyzed by means of the following parameters : residual echo level, complexity of computations in the case of finite computational accuracy and speed of convergence. These algorithms are the well-known gradient algorithm and the one which uses the sign of the error. The authors show that adding a controlled noise in the sign-algorithm makes its residual echo and the length of its binary words behave like those of the gradient algorithm. The sign algorithm total complexity is less than the gradient algorithm one. The problem of the lower speed of convergence can be solved by means of simple initialization techniques. Theoretical results are illustrated through computer simulations.     

The Effects of Large Interference on the Tracking Capability of Digitally Implemented Echo Cancellers

Adaptive mean-square-tapped-delay-line echo cancellers for voice applications are conventionally designed to stop adjustment during periods of "double-talking", i.e., when a large informationbearing signal is present along with the echo signal to be cancelled. Continuous adaption is, however, desirable in full-duplex, two-wire data transmission where the periods of double-talking are so long that the echo channel may vary. We presume that the tap weights of an echo canceller have converged during a training period free of double talking, and address the problem of subsequent echo-canceller tap adjustment via the estimated-gradient algorithm in the presence of double talking. In the estimated-gradient algorithm the tap increment should be proportional to the product of the residual echo and the tap voltage. However, when double talking occurs the residual echo can only be estimated. For an idealized double-talking model, it is demonstrated, from infinite-precision considerations, that use of the memoryless maximum-likelihood estimate of the residual echo is nearly equivalent to abrupt reduction of the step size of the adjustment algorithm when double-talking begins, and could provide an automatic mechanism for recognizing double-talking. Unfortunately, the response of a digitally implemented canceller to a sharply reduced step size can be a deterioration in performance. In fact, the use of an exceedingly small step size during periods of doubletalking may lead to a cancellation error considerably larger than that predicted by coefficient precision. It is demonstrated how averaging the estimated gradient can significantly decrease the mean-squared tap error during periods of double talking. To a first approximation, the tap-weight error can be reduced by a factor proportional to the averaging interval, with an equivalent decrease in tracking capability.     

A New Normalized Subband Adaptive Filter Algorithm with Individual Variable Step Sizes

Proposed is a novel variable step size normalized subband adaptive filter algorithm, which assigns an individual step size for each subband by minimizing the mean square of the noise-free a posterior subband error. Furthermore, a noniterative shrinkage method is used to recover the noise-free priori subband error from the noisy subband error signal. Simulation results using the colored input signals have demonstrated that the proposed algorithm not only has better tracking capability than the existing subband adaptive filter algorithms, but also exhibits lower steady-state error.     

An intensity independent fixed valued impulse noise detector for image restoration

Most of the impulse noise detectors used for detection of fixed valued impulse noise are effective only for salt and pepper or a band type noise occurring at the extreme ends of the allowed range of intensity levels. The performance of these detectors deteriorates drastically when fixed valued impulses occur anywhere within the allowed gray scale. In this paper, an impulse detection scheme is proposed which can effectively detect all types of fixed valued impulse noise and also differentiates between noisy and noise-free pixels of identical intensity levels. The improved performance of the proposed method is verified through extensive simulations for various fixed valued impulse noise models.     

VoIP回声消除器设计及算法研究

本文提出了一种与线性预测编解码器相结合的新声学回声消除器,由去相关可变步长的NLMS自适应算法和基于回声路径失配方差的双端通话检测算法所组成.Matlab仿真结果表明,与Gordy所提出的回声消除算法相比,本文提出的算法在双端通话和回声路径改变时判别更准确,收敛速度更快;在收敛状态时,ERLE值平均提高了15dB,失调误差平均降低了10dB,具备更好的回声消除性能.     

A covariance matrix shrinkage method with Toeplitz rectified target for DOA estimation under the uniform linear array

A covariance matrix shrinkage method is proposed to make an improvement of the direction of arrival (DOA) estimation under a uniform linear array in a scenario where the number of sensors is large and the sample size is relatively small. The main contribution is that we provide a shrinkage target with Toeplitz structure and deduce a closed-form estimation of the shrinkage coefficient. The closed-form and the expectation of the shrinkage coefficient estimate are calculated based on the unbiased and consistent estimates of the trace and moments of a Wishart distributed covariance matrix. The statistical property of the shrinkage coefficient estimate is discussed through theoretical analysis and simulations, which demonstrate the shrinkage coefficient estimate can ensure that the proposed covariance matrix estimate is a good compromise between the sample covariance matrix (SCM) and the target. The root-mean-square-error (RMSE) simulations of DOA estimation show that the proposed method can improve the multiple signal classification (MUSIC) DOA estimation performance in the case of low signal-to-noise ratio (SNR) with small sample size, and also can provide a satisfactory performance at high SNR.     

Monochromatic-Based Method for Impulse Noise Detection and Suppression in Color Images

A new method to detect and reduce the impulse noise in color images is presented in this paper. The method consists of two stages: detection and filtering. Since each of the individual channels (components) of the color image can be considered as a monochrome image, both stages are applied to each channel separately, and then the individual results are combined into one output image. The corrupted pixels are detected in the first stage based on a proposed innovative switching technique. The noise-free pixels are copied to their corresponding locations in the output image. In the second stage, average filtering is applied only to those pixels which are determined to be noisy in the first stage, and only noise-free pixel values are involved in calculating this average. The size of the sliding window depends on the estimated noise density and is very small even for high noise densities. The proposed method is effective in noise reduction while preserving edge details and color chromaticity. Simulation results show that the proposed method outperforms all the tested existing state-of-the-art methods used in digital color image restoration in both standard objective measurements and perceived image quality.     

Accelerated gradient with optimal step size for second-order blind signal separation

This paper proposes an algorithm for the second-order blind signal separation problem with convolutive mixtures. An iterative first order gradient method based on the accelerated gradient is developed for solving the optimization problem. For each search direction, the question becomes how to effectively calculate the optimal step size in each iteration. Here, we propose an efficient algorithm for obtaining the step size by first reformulating the objective function as a fourth order polynomial in terms of the step size, where the polynomial coefficients are required to be calculated only once per iteration. An optimal step size search procedure using the Newton’s method is developed with the step size is efficiently obtained for each iteration. Simulation results in a simulated room environment and a real environment show that the proposed algorithm converges faster than the existing methods with a lower number of iterations and a lower computational complexity. In addition, the proposed algorithm can separate the speech signals and reduce the background noise simultaneously.     

A new eigenstructure method for sinusoidal signal retrieval inwhite noise: estimation and pattern recognition

A new approach, in a framework of an eigenstructure method using a Hankel matrix, is developed for sinusoidal signal retrieval in white noise. A closed-form solution for the singular pairs of the matrix is defined in terms of the associated sinusoidal signals and noise. The estimated sinusoidal singular vectors are applied to form the noise-free Hankel matrix. A pattern recognition technique is proposed for partitioning signal and noise subspaces based on the singular pairs of the Hankel matrix. Three types of cluster structures in an eigen-spectrum plot are identified: well-separated, touching, and overlapping. The overlapping, which is the most difficult case, corresponds to a low signal-to noise ratio (SNR). Optimization of Hankel matrix dimensions is suggested for enhancing separability of cluster structures. Once features have been extracted from both singular value and singular vector data, a fuzzy classifier is used to identify each singular component. Computer simulations have shown that the method is effective for the case of “touching” data and provides reasonably good results for a sinusoidal signal reconstruction in the time domain. The limitations of the method are also discussed     

遗传算法的噪声干扰数字图像分类性能评价

针对现实中各种噪声干扰的数字图像识别分类的问题,提出了基于遗传算法优化的BP神经网络和支持向量机神经网络两种方案,先在无噪声干扰情况下建模,然后加入人工噪声模拟现实中的噪声干扰。结果表明,遗传算法优化后的支持向量机网络方案具备更好的抗噪声干扰能力,在噪声干扰数字图像分类中具有更高应用价值。     

Acoustic echo-canceler using the FBAF algorithm

An acoustic echo-canceler for teleconferencing systems is realized based on the frequency bin adaptive filtering (FBAF) algorithm. In the FBAF algorithm, each frequency bin does an independent adaptive filtering, so that parallel processing can be used to increase the throughput of the system. Hardware size can be reduced to about 25% of the FIR time domain adaptive filter (TDAF) requirement. The realized echo canceler allows a comfortable conversation with only 8 ms of delay. The hardware prototype contains 12 VSP chips and one DSP chip, An ERLE (echo return loss enhancement) of 30 dB was achieved using this prototype hardware for an echo reverberation path with 260 ms delay. An efficient method for normalizing the convergence factor of the FBAF algorithm with a correlated input signal is given that speeds up the convergence rate. The performance is shown by computer simulation     

Performance analysis of an echo-cancellation arrangement thatcompensates for frequency offset in the far echo

Analytical and experimental results are presented for the performance of one echo canceller arrangement. It consists of a data-driven echo canceller having a so-called cross-coupled structure, which is followed by a rotator and a phase-locked loop (PLL). A cross-coupled echo canceller structure without a PLL is analyzed first. Expressions for speed of convergence and achievable echo-return-loss enhancement (ERLE) in the presence of frequency offset are derived. These results are compared in previously published results for a noncross-coupling echo canceller structure. Specifically, it is shown that the cross-coupled structure converges twice as fast as the noncross-coupled structure and provide an achievable ERLE that is about 6 dB better. The joint adaptation of the echo canceller and the PLL is then studied. It is shown that it is always possible to choose design parameters for the echo canceller which are consistent with adaptation requirements under double-talking conditions, provided that the PLL is properly engineered. The sensitivity of the performance of PLL to the power level of the far echo, as well as possible solutions to this problem, are discussed     

小步进频率合成器的设计

回顾了三种基本的小步进频率合成器设计方法的优缺点。介绍了一种特殊的小步进频率合成器的设计方法，即采用两个大步进频率的单环锁相电路混频，两者步进频率的差较小为r，就能获得输出为小步进频率(为r)的合成器，并给出了相应的理论依据和计算。只要合理设置频率，规避互调分量的影响，就能使合成信号保持大步进频率单环锁相电路较低相位噪声、较短跳频时间和较低杂散信号的特性，而且合成原理简单，几乎不需要电路调试。     

Double-Talk-Robust Prediction Error Identification Algorithms for Acoustic Echo Cancellation

The performance of an acoustic echo canceller may be severely degraded by the presence of a near-end signal. In such a double-talk situation, the variance of the echo path estimate typically increases, resulting in slow convergence or even divergence of the adaptive filter. This problem is usually tackled by equipping the echo canceller with a double-talk detector that freezes adaptation during near-end activity. Nevertheless, there is a need for more robust adaptive algorithms since the adaptive filter's convergence may be affected considerably in the time interval needed to detect double-talk. Moreover, in some applications, near-end noise may be continuously present and then the use of a double-talk detector becomes futile. Robustness to double-talk may be established by taking into account the near-end signal characteristics, which are, however, unknown and time varying. In this paper, we show how concurrent estimation of the echo path and an autoregressive near-end signal model can be performed using prediction error (PE) identification techniques. We develop a general recursive prediction error (RPE) identification algorithm and compare it to three existing algorithms from adaptive feedback cancellation. The potential benefit of the algorithms in a double-talk situation is illustrated by means of computer simulations. It appears that especially in the stochastic gradient case a huge improvement in convergence behavior can be obtained