Muthuswamy J, Thakor N V
Department of Biomedical Engineering, Johns Hopkins University, Baltimore, MD 21205, USA.
J Neurosci Methods. 1998 Aug 31;83(1):1-14. doi: 10.1016/s0165-0270(98)00065-x.
This paper reviews some novel spectral analysis techniques that are useful for neurological signals in general and EEG signals in particular. First, some drawbacks and limitations of the commonly used Fast Fourier transforms (FFTs) are presented, and then alternative algorithms are outlined. An auto-regressive (AR) modeling based spectral estimation procedure is presented to overcome the problems of lower resolution and 'leakage' effects inherent in the FFT algorithm. For signals which are transient in nature or rapidly time-varying, two alternative algorithms are presented. The first is an adaptive AR parameter estimation algorithm and the second is a wavelet based time-frequency representation algorithm. Finally, a Spectral Distance measure and the Itakura distance measure are presented to quantify the differences between the spectra of two signals in a succinct manner. The application and performance of all the algorithms is illustrated using electroencephalograms (EEGs) recorded in animals during hypoxic asphyxic injury to brain.
本文综述了一些新颖的频谱分析技术，这些技术一般对神经信号有用，尤其对脑电图（EEG）信号有用。首先，介绍了常用快速傅里叶变换（FFT）的一些缺点和局限性，然后概述了替代算法。提出了一种基于自回归（AR）建模的频谱估计程序，以克服FFT算法固有的分辨率较低和“泄漏”效应问题。对于本质上是瞬态或快速时变的信号，提出了两种替代算法。第一种是自适应AR参数估计算法，第二种是基于小波的时频表示算法。最后，提出了频谱距离度量和伊塔库拉距离度量，以简洁的方式量化两个信号频谱之间的差异。使用动物在脑缺氧性窒息损伤期间记录的脑电图（EEG）说明了所有算法的应用和性能。
