Spectral analysis of EEG responses.

The techniques used and the results obtained in a spectral analysis of two specific responses in the human electroencephalogram are presented in this paper. The purposes are to show how the techniques may be applied to the necessarily short lengths of EEG data and to illustrate these techniques and the useful results obtained by relevant examples. The necessary data-processing procedures and precautions for transforming from the time to frequency domain are presented in a tutorial fashion. The importance of augmenting zeros, choice of the most appropriate data window and pretransformation of the data to avoid the combined effects of energy loss and low frequency content biasing caused by windowing is explained. The pros and cons of the tapered-cosine (Tukey) and Kaiser-Bessel windows are illustrated. The usefulness of applying certain statistical tests, which are based on a physical model of the responses, to the harmonic components of the responses is demonstrated. Thus a comparison is made between the features of auditory evoked responses and of the contingent negative variation, and the usefulness of predictive statistical diagnosis in differentiating between subject groups is illustrated by application to normal subjects and Huntington's chorea patients.