Application of State-Space Modeling to instantaneous independent-component analysis

In this paper, we design an algorithm for decomposing multivariate electroencephalographic (EEG) time series into independent components, based on Independent-Component Analysis (ICA) and State-Space Modeling (SSM). We aim at combining the strong aspects of both methods: ICA provides an initial model for SSM which is then further optimized by maximum-likelihood. We also propose an approach for augmentation of the state space by extracting additional components from the data prediction errors. The estimate of the mixing matrix provided by ICA is excluded from optimization. Practical application of the proposed algorithm is demonstrated by an example of the analysis of EEG data recorded from an epilepsy patient.