Engelken E J, Stevens K W, Enderle J D
Armstrong Laboratory/NG, Brooks AFB, TX 78235.
Biomed Sci Instrum. 1991;27:163-70.
An adaptive nonlinear digital filter has been designed for the analysis of an eye-movement signal called nystagmus. Nystagmus is a bi-phasic signal consisting of a sequence of tracking eye movements called "slow-phase" interspersed with brief, high-velocity refixation movements called "fast-phase." The objective of the analysis is to separate the nystagmus signal into its fast- and slow-phase components. Specifically, the goal is to produce an evenly sampled estimate of slow-phase velocity (SPV) and an estimate of the peak fast-phase velocity. Classically this has been done using pattern recognition methods that exploit the fact that the fast-phase is a relatively short duration, high-velocity movement compared to the slow-phase. Unfortunately, these velocity and duration differences do not reliably separate the slow- and fast-phases under all conditions, especially when the signal is noisy. We have designed and built an adaptive nonlinear digital filter that easily outperforms the more complex pattern recognition algorithms. This new filter, called an Adaptive Asymmetrically Trimmed-Mean (AATM) filter, works under the assumption that, on the average, the eyes spend more time in slow-phase than in fast-phase. Thus, in any given data segment, most of the data samples are slow-phase samples. By analyzing the amplitude distribution of the data samples in the segment we can determine which of these samples are slow-phase. We used computer generated nystagmus signals contaminated with 3 levels of noise to evaluate the filter. The filter parameters were then optimized using Monte Carlo procedures producing an extremely robust analysis method.
一种自适应非线性数字滤波器已被设计用于分析一种称为眼球震颤的眼动信号。眼球震颤是一种双相信号，由一系列称为“慢相”的跟踪眼动序列组成，其间穿插着称为“快相”的短暂、高速的重新注视运动。分析的目的是将眼球震颤信号分离为其快相和慢相成分。具体来说，目标是生成慢相速度（SPV）的均匀采样估计值和快相峰值速度的估计值。传统上，这是通过模式识别方法来完成的，这些方法利用了快相相对于慢相是持续时间较短、速度较高的运动这一事实。不幸的是，这些速度和持续时间差异在所有情况下都不能可靠地分离慢相和快相，尤其是当信号有噪声时。我们设计并构建了一种自适应非线性数字滤波器，它很容易超越更复杂的模式识别算法。这种新的滤波器称为自适应非对称修剪均值（AATM）滤波器，其工作假设是，平均而言，眼睛在慢相花费的时间比在快相花费的时间更多。因此，在任何给定的数据段中，大多数数据样本都是慢相样本。通过分析该段中数据样本的幅度分布，我们可以确定哪些样本是慢相样本。我们使用受3种噪声水平污染的计算机生成的眼球震颤信号来评估该滤波器。然后使用蒙特卡罗程序对滤波器参数进行优化，从而产生一种极其稳健的分析方法。
