Group Blind Source Separation (Gbss)

A new group blind source separation (GBSS) method is introduced and is based on the group structure of the spectral properties of the latent sources. The hidden sources are assumed to be stationary time series. Their spectral density functions will be estimated directly from the sensors or mixed signals according to the specified or known group structures, which can be furthur described by space or time or both. Our results will be compared with a group independent component analysis method (GroupICA) based on some popular ICA algorithms. This will be carried out using simulated and real human brain data.