利用合同变换矩阵的时间结构信号盲分离算法

Blind separation algorithm for sources with temporal structure based on contract transformation matrix

在信号处理领域中,很多源信号具有时间结构,即时间信号或时间序列。针对这种具有时间结构的盲源分离问题,本文利用源信号与观测信号的自相关矩阵之间存在的一种合同关系,即分离矩阵相当于将一个观测信号的自相关矩阵化为对角矩阵的合同变换矩阵,基于这种合同变换矩阵的思想,改进了时间结构独立成分分析算法,将求分离矩阵转化为求各个时延下合同变换矩阵集合的交集。最后,通过MATLAB对仿真模拟源信号及真实语音信号两个实例进行分离实验,验证了该分离算法的有效性。

In the field of signal processing, there are many source signals with temporal structure, i.e. time signals or time series. In the procedure of dealing this blind source separation problem, there is a contractual relationship between the autocorrelation matrix of the source signals and the autoeorrelation matrix of the observed signals under the linear instantane- ous mixing model, in other words, the separation matrix is a contract transformation matrix which transforms the autoeorre- lation matrix of the observed signals into a diagonal matrix, which is corresponding to independent sources. Based on this relationship, we improve the Independent Component Analysis （ICA） algorithm of time signals, that is, the intersection of the contract transformation matrix sets under each delay is the required separation matrix. Finally, the simulation results for analog source signals and real sound signals show that the new algorithm is valid and efficient.