用慢特征分析算法实现水声信号盲分离

Blind source separation of underwater acoustic signals by using slowness feature analysis

在常规的水声信号盲处理研究中,通常都是用独立成分分析算法分离线性混合信号,而对于较复杂的非线性混合信号,独立成分分析算法无能为力。针对这种情况,提出将慢特征分析(Slow Feature Analysis,SFA)算法应用于水声信号非线性盲源分离领域。一般而言,对源信号做非线性混合变换后输出混合信号较源信号变化较快,而采用SFA算法可以从复杂的非线性混合信号中提取出变化缓慢的信号,通过仿真实验,分别对简单信号和复杂水声信号的非线性混合信号进行分离,通过将源信号与分离信号对比,发现SFA算法输出信号与源信号高度相似,验证了SFA算法在非线性盲源分离领域应用的有效性和可行性。

In conventional blind underwater acoustic signal processing, the independent component analysis algorithm is often used to separate linear mixed signals. However, for the more complex nonlinear mixed signal, the independent component analysis algorithm is helpless. To solve this problem, this article applies slow feature analysis to blind un- derwater acoustic signal processing. In general, the nonlinear mixed signal varies faster than the source signal does, and SFA algorithm can extract slowly varying features from complex nonlinear signals. Through simulation experiment, the nonlinear mixed signals of simple signals and complex underwater acoustic signals are separated. By comparing the source signals and the separated signals, it is found that the output signals of SFA correlate to the source signal highly. It proves that SFA is effective and practicable in the field of nonlinear blind source separation application.