ARMA model estimation for EEG using canonical correlation analysis

The EEG signal is modeled as an autoregressive moving-average (ARMA) process. The performance of the canonical correlation analysis algorithm to estimate the order of the AR and MA polynomials for real and simulated EEG signals is investigated. It is shown that for records of 5-s duration or more the algorithm gives good and consistent estimates of the model order and AR coefficients and is insensitive to the location of the poles relative to the unit circle. The method is also insensitive to the percentage energy in the component waves.<
>