A Fast Independent Component Analysis Algorithm for Improper Quaternion Signals

An extension of the FastICA algorithm is proposed for the blind separation of both Q-proper and Q-improper quaternionvalued signals. This is achieved by maximising a negentropy-based cost function, and is implemented using the Newton method in the augmented quaternion statistics framework. It is shown that the use of augmented statistics and the associated widely linear modeling provide theoretical and practical advantages over standard models. Simulations using both benchmark and real-world signals support the approach. Index Terms Independent Component Analysis (ICA), augmented quaternion statistics, quaternion widely linear modelling, quaternion noncircularity, quaternion blind source separation