基于参数模型和独立分量分析的事件相关诱发电位单次提取

Extraction of Single-trial Event-related Potentials by Means of ARX Modeling and Independent Component Analysis

我们针对脑电事件相关电位(ERP)这种信噪比极低的信号检测问题,提出了两种ERP信号单次提取方法,能非常有效地同时去除自发脑电、眼动伪迹和工频噪声三种常见噪声。(1)首次对自发脑电、眼动伪迹和工频噪声这三种常见成分连同事件相关电位同时进行ARX建模,利用基于最小二乘(LS)的ARX算法进行参数辨识获得提取结果;(2)利用独立分量分析,采用FastICA算法进行事件相关电位的提取。明确指出ICA分解的一些重要分解特性及其内在机理,针对实际情况对FastICA算法进行了改进,实现了分解结果对ERP成分的自适应映射。数值仿真实验结果表明两种方法均有较高的信号分解提取能力。

The present paper focused on the extraction of event-related potentials on a single sweep under extremely low S/N ratio. Two methods that can efficiently remove spontaneous EEG, ocular artifacts and power llne interference were presented based on ARX modeling and independent component analysis （ICA）. The former method applied ARX model to the measured compound signal that extensively contained the three kinds of ordinary noises mentioned above, and used ARX algorithm for parametric identification. The latter decomposed the signal by means of independent component analysis. Besides, some of ICA＇s important decomposing characters and its intrinsic causality were pointed out definitely. According to the practical situation, some modification on FastlCA algorithm was also given, so as to implement auto-adaptive mapping of decomposed results to ERP component. Through simulation, both the two ways are proved to be highly capable of signal extraction and SIN ratio improving.