The performance of the least mean square algorithm will degrade significantly when the background noise is impulsive non-Gaussian distributed. However the least mean p-norm algorithm can still work well under alpha stable noise conditions. In this paper, we present a sweep-by-sweep adaptive evoked potential estimator which utilizes a radial basis function neural network with a time index series input under impulsive noise conditions. This estimator uses matrix form for weights updating and does not need to design the reference signal. The least mean p-norm algorithm cannot work well when the value of alpha dynamically changes. To overcome this shortcoming, we propose a kind of improved algorithm, namely signed direct adaptive algorithm.Experimental results show that the improved one can work well when alpha dynamically changes. The correlation coefficient between estimated signal and underlying source signal is still greater than 0.9 even if mixed signal-to-noise ratio is - 12 dB, and we can draw the conclusion that signed direct adaptive algorithm is one kind of robust adaptive estimator.%诱发电位(EP)观测信号中的背景噪声经常呈现出某种非高斯脉冲特性,使基于高斯假设的最小均方自适应算法性能明显退化,而适用于α稳定分布假设的最小平均P范数算法仍能较好地工作.借鉴该思想,使用以时间索引序列为输人的径向基神经网络,实现脉冲噪声下的EP信号自适应估计,给出矩阵形式的权值更新公式,逐扫描地完成自适应估计,且无需设计参考信号.针对当α动态变化时最小平均P范数算法性能变差的不足,提出基于符号函数直接自适应的改进算法用于EP信号估计.实验结果表明,改进后的算法可以在α动态变化时很好地跟踪EP信号,即使在很低的混合信噪比时(-12 dB),估计信号与EP源信号的相关系数仍在0.9以上,是一种在α稳定分布噪声下具有良好韧性的EP信号自适应估计算法.
展开▼
