Multichannel Autoregressive Order Selection in Practice

Multichannel time series analysis takes the same model order and model type for the different vector signals involved. Selection criteria have been developed to select the best order to predict the different components of the vector simultaneously. The prediction of single channels might require a different order or type for the best accuracy of each signal. That can become a problem in multichannel analysis if the individual signals have completely different model orders. Univariate and multichannel spectra are not similar then. Furthermore, the selected order may vary in practice with the number of signals in a multichannel vector. A turbulence example shows the results of order selection, and the consequences in estimating the coherency between vector signals with the dimensions two and five.