ARMA order selection for EEG-an empirical comparison of three order selection algorithms

The performance of three ARMA (autoregressive moving-average) order estimation algorithms, canonical correlation analysis, S-array, and Franke algorithm, on simulated and real electroencephalogram (EEG) signals is presented. It is shown that the S-array always correctly indicates the AR order but makes incorrect estimates of the MA order. The canonical correlation method identifies the model order correctly for simulated data and overestimates the AR order on real data. The Franke algorithm is shown to perform poorly in comparison to the other algorithms