Abstract:
A complex-valued radial basis function neural network RBF is proposed for digital communications channel equalization. Performances are directly related to the clustering centers estimations. For this aim different competitive learning algorithms are developed. The network has complex centers and connection weights, but the nonlinearity of its hidden nodes remains a real-valued function. The RBF network is able to generate complicated nonlinear decision regions or to approximate an arbitrary nonlinear function in complex multidimensional space.
