Nonlinear multilayer principal component type subspace learning algorithms

A hidden layer is introduced into nonlinear principal component type learning algorithms. The algorithms are derived from nonlinear optimization criteria. Both subspace type and hierarchical versions are considered. The algorithms are tested in context with harmonic retrieval and directions-of-arrival estimation problems using impulsive and colored noise. Some of the nonlinear algorithms have interesting signal separation properties, i.e., the neurons become sensitive to independent sinusoidal signals