信号分离

Then we have compared the directional characteristic of the basic wavelets; according to the property of impulse function, we have simplified the formula of inverse transformation.

针对二维信号处理中的波场分离，引入了方向性小波的概念，将其用于信号波场分离，对比了三种基本小波的方向特性；根据函数的性质，提出更为简单的二维连续小波反变换公式。

Considering the real speech mixture model could be nonlinear-convolution mixture model, the author presents a new nonlinear speech separation algorithm by utilizing the Finite Impulse Response neural network and the GMM pdf estimation, based on the feedback architecture and the Maximum Likelihood principle.

考虑到真实多通道语音分离系统的输入混合语音有可能是不同语音源信号的非线性混合而不是简单的线性混合，基于反馈结构，采用GMM概率密度函数估计方法，作者提出了一种采用有限冲激响应神经网络的非线性、卷积混合盲信号处理语音分离批处理新算法。

The ICA is a kind of signal processing method based on high order statistic, it can recover the source signals from the linearly mixed signals, but it has disadvantages in processing non-linearity signals, and the KICA, which is based on nonlinear function space, can solve the problem effectively. Comparing with current ICA algorithms, the KICA is notable for its flexibility and robustness.

独立分量分析是近几年发展起来的基于信号高阶统计特性的分析方法，它是指从多个源信号的线性混合信号中分离出源信号的技术，但目前的算法在处理非线性变化的信号时还有一定的局限，而基于非线性函数空间的ICA方法－KICA，即核独立成分分析，可以解决这一问题。

The technical ofblind signal separation is mainly discussed in this dissertation: including the theory and themethod. Then a multi-sensor intelligent data processing system is established to solve theproblem involved above. And this is belong to the pels fusion and the characteristic fusion ofinformation fusion. The research content and contributions of the author are summarized asfollows:(1) The pre-processing procedure of blind signal separation is mainly discussed in this chapter.

本文主要研究了信号盲分离技术的理论与方法，并在此基础上研究了一个多传感器数据智能处理系统用于解决多传感器信息融合系统中当传感器观测信号为实际目标信号的混合信号时，如何从观测信号中提取目标源信号的问题，它属于信息融合功能层次中的像素级融合和特征级融合。

An on-line blind source separation algorithm of non-stationary mixed signals is presented. The algorithm uses the multiple time-delayed decorrelation and natural gradient rule to obtain the stationary point of the cost function, only a small number of observed samples is needed to minimize the Frobenius norm of the correlation matrices, so the computing cost is low and continuous separation calculation could be realized by on-line operation.

利用一种基于多时延解相关准则的代价函数和自然梯度原则，分别推导出非平稳信号在瞬态线性混合及卷积混合情况下的在线盲分离算法，仅通过对有限个时延样本的相关矩阵进行Frobenius范数最小化运算得到分离矩阵，计算工作量小，无需对样本进行分块处理，可实现对信号的连续跟踪操作。

Control layer and parameter layer represent the available number of rows and the concrete value of the available rows of separation matrix respectively. Non-Gaussian was used to measure the independence of the stochastic signals. The kurtosis was maximized/minimized by searching a linear combination of observation signals, and then the independent variables were separated.

用控制层、参数层分别表示分离阵有效行数和有效行数的具体数值，将非高斯性作为随机信号相互独立的度量，通过寻求观测信号的一个线性组合，使其峭度最大化或最小化来分离独立分量。

Only an arbitrary single pulse among pulse train with temporal separation of about 1 ps was trapped by soliton pulse,the trapped pulse was blue-shifted and the optical spectra of trapped pulse was distinctly separated from those of the untrapped signal pulses.The trapped pulse can be easily picked off using a wavelength filter such as a fiber Bragg grating.

通过数值求解光子晶体光纤中脉冲传播满足的耦合非线性薛定谔方程，发现用孤子脉冲可俘获信号脉冲串中的任何一个脉冲，被俘获的信号脉冲的中心波长明显蓝移，在频域上和其他信号脉冲分离开来，于是让信号脉冲串在输出端通过布喇格光纤光栅，被俘获的信号脉冲将被过滤掉。

Convolved audio signals can be separated by frequency-domain blind source separation. Complex independent component analysis is performed in each frequency bin of the mixed signal, but the permutation inconsistency and scaling arbitrary are involved.

混合语音信号可以使用盲分离频域解法，对观测信号在每一个频点分别进行复值独立分量分析算法来解混并得到分离信号，但带来了幅值和次序不定问题。

The signals during transformer impulse test are considered to be evolving in time, also is non-stationary signals, so we can design a software algorithm for noise reduction based on discrete Gabor transform, and the basic principle is: The Gabor coefficients of Effective signals during transformer impulse test is centralized in the time frequency range, and the peak-to-peak value is big, whereas the noise Gabor coefficient distribute in the entire time frequency range equally, the peak-to-peak value is small, we can realize the desired signal and the noise separated through frequency mask function.

考虑到冲击试验中的信号都是时变信号，也就是非平稳信号，因此设计了一种基于离散GABOR变换的软件滤波算法对冲击信号进行滤波，其滤波理论依据是：有效信号的Gabor系数在时频域分布比较集中，幅值比较大，噪声的Gabor系数均匀分布于整个时频域，幅值较小，通过构造时频掩模函数来实现有效信号与噪声的分离。

At the same time, pattern recognition has been used for abstracting characteristics、predicting the life and evaluating the properties of engine. This research contributes toward global diagnosis for the diesel engine in eight ways.(1) Put forward"the Arithmetic of distinguish the period automatically using auto-correlation method";(2) Separate the waves of continuous knocks using WV distribution;(3) Propose"the logarithmic spectrum approximate method"to analysis the response of the knock between the steel and steel;(4) Discuss the relationship between the time-frequency distribution and timescale distribution in detail, and give the principle of classification;(5) Use the wavelet method to analysis the pressure signal of exhaust、suction;(6) Use the shape of vibration signal to analyse the working situation of cylinder;(7) Use the pattern recognition to select the good samples and abstract the effective characteristics;(8) Develop the model of total properties evaluation and life prediction.

本文在广泛的、深入研究的基础上，应用多种信号处理手段来分析处理数据，同时利用模式识别方法对故障的特征提取、寿命预测以及性能评价作了探讨，本文进行了以下几项工作：（1）提出了基于自相关自动周期识别算法的柴油机振动信号预处理策略；（2）提出了基于WVD分布的连续撞击响应信号的分离策略；（3）提出了基于对数谱拟合方法的钢—钢撞击响应求解模型；（4）提出了基于波形分析的非正常燃烧振动信号的诊断模型；（5）提出了基于小波细节信息的进气压力信号和曲轴箱压力信号的诊断模型；（6）提出了基于小波近似的喷油正时信号中燃烧位置的自动定位；（7）提出了基于统计模式识别方法的故障诊断知识的自学习与自组织策略；（8）提出了基于特征压缩思想的柴油机总体性能评价及寿命预测模型。

Finally, according to market conditions and market products this article paper analyzes the trends in the development of camera technology, and designs a color night vision camera.

最后根据市场情况和市面上产品的情况分析了摄像机技术的发展趋势，并设计了一款彩色夜视摄像机。

Only person height weeds and the fierce looks stone idles were there.

只有半人深的荒草和龇牙咧嘴的神像。