Data length effects on the coherence estimate of EEG

The coherence functions between EEGs (electroencephalograms) collected from the hippocampal subfield CA1 and dentate is reexamined by conventional (periodogram) and autoregressive (AR) approaches to study the effects of different data lengths on the performance of these approaches and the variation of the resultant magnitude squared coherence (MSC) and phase coherence estimates. The MSC and phase coherence obtained from the autoregressive method are smoother and demonstrate lower standard deviation than the results obtained by the classical approach. In general, the standard deviation of the MSC and the phase coherence decreases as data length increases, and the standard deviation of the AR method is less than that of the periodogram. The results indicate that for applications in which computational speed is not critical, the multichannel autoregressive model provides a coherence estimate that is smoother and exhibits lower standard deviation than the conventional periodogram approach