Online Detection of the Nature of Complex-Valued Signals

A novel method for on-line tracking of the changes in the nature of a complex-valued signal is proposed. This is achieved by analysing the time variation of the mixing parameter within a hybrid complex-valued nonlinear adaptive filter. The proposed hybrid filter consists of a combination of split- and fully-complex nonlinear gradient descent algorithms, whose outputs are mixed in a convex manner. A learning algorithm for this scheme is derived and the potential of such an approach for tracking of signal modality changes is highlighted. The potential of the proposed approach is supported by simulations on both a synthetic benchmark signal and on real-world radar data.