A blind source separation framework for detecting CPM sources mixed by a convolutive MIMO filter

This paper deals with blind separation of convolutive mixtures of continuous phase modulated (CPM) sources. The main difficulty lies in the fact that CPM sources are non-linear (and hence non i.i.d.) sources. The problem is addressed through the general formulation of blind source separation (BSS). The separation method consists in iterative constrained optimizations of criteria depending on the fourthorder statistics. We prove the validity of the considered contrast functions for the extraction of one source. A local study then allows us to show that the optimization is free of spurious local maxima at each step and that it is possible to alleviate the error accumulation problem by using an unconstrained post-optimization technique. After separation is achieved, the emitted symbols are estimated, based on recent results concerning CPM equalization. Finally, simulations illustrate the validity of the method.