自适应变步长盲源分离算法的研究

发布时间：2019-04-28 17:43

【摘要】：信号处理作为一种强大的获得信号的方法,可以快速有效地得到所需要的信号。但是,不可避免地,通常采集到的信号都有着其他不希望看到的成分。如果在处理之前,可以知道一些信号的信息以及它们是怎样混合的,那么问题变得相对简单一些。但是,其实学者们未能得到所有混合后的信号特征以及数目,此时,盲源分离作为一种功能强大的方法取得人们的信任,它可以不需要知道信号的任何特征而逐个分离出可能的源信号,并且在此方向取得了很高的评价。自适应算法的优劣与否直接与步长的变化息息相关。计算量最小的做法是采用固定步长,但是,步长稍大时,无法得到精确的分离信号,如果采用很小的步长,则使得收敛速度非常慢,有时甚至因为已经到了收敛阶段,信号还未得到分离而永远无法分离出源信号。因此,固定步长虽然运行简单,操作方便,但是无法同时满足稳定性好和收敛快这两个要求。本文主要在自然梯度盲源分离方法的基础上进行深入分析,引入了模拟退火思想,将过程分为两个阶段,第一阶段借鉴模拟退火算法采用固定步长减小计算量且加快收第二阶段用分离状态作为指数部分有目标地控制步长变化。通常盲源分离为了简化计算,需要对信号进行预处理,白化是较常被用到的。这样做简化了算法,使得算法的复杂度降低将近一半,但同时也限制了算法的应用范围,可能会导致算法不具有等变化性。而且很容易受到混合矩阵的影响,可能会使分离出的信号很大程度上偏离源信号。因此,本文在盲源分离的过程中加入正交约束,不仅可以达到预白化的效果而且还可保留盲源分离的等变化性质,分离效果更稳定。
[Abstract]:As a powerful method to obtain the signal, signal processing can get the desired signal quickly and effectively. But inevitably, the signals that are usually collected have other components that they don't want to see. The problem becomes relatively simple if information about some signals and how they are mixed can be known before processing. However, in fact, scholars have not been able to obtain all the mixed signal characteristics and numbers, at this time, blind source separation as a powerful method to gain people's trust, It can separate the possible source signal one by one without knowing any characteristics of the signal, and has obtained a high evaluation in this direction. Whether or not the adaptive algorithm is good or bad is closely related to the change of step size. The minimum amount of calculation is to adopt a fixed step size, but when the step size is slightly larger, it is impossible to obtain accurate separation signals. If a small step size is used, the convergence rate will be very slow, sometimes even because it has reached the stage of convergence. The signal has not yet been separated and will never be able to separate the source signal. Therefore, although the fixed step size is simple to run and convenient to operate, it can not meet the requirements of good stability and fast convergence simultaneously. In this paper, based on the analysis of the natural gradient blind source separation method, the idea of simulated annealing is introduced, and the process is divided into two stages. In the first stage, using simulated annealing algorithm for reference, the fixed step size is used to reduce the computational complexity and the second stage is used as the exponential part to control the step size change with the separation state as the exponential part. In order to simplify the calculation, blind source separation usually needs to pre-process the signal, and whitening is more commonly used. This simplifies the algorithm, reduces the complexity of the algorithm by nearly half, but also limits the scope of application of the algorithm, which may lead to the algorithm does not have equal variability. Moreover, it is easy to be influenced by the mixed matrix, which may cause the separated signal to deviate from the source signal to a large extent. Therefore, adding orthogonal constraints in the process of blind source separation can not only achieve the effect of pre-whitening, but also retain the variable properties of blind source separation, and the separation effect is more stable.
【学位授予单位】：太原理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TN911.7

【相似文献】