A systolic array for nonlinear adaptive filtering and pattern recognition

A systolic array for multidimensional fitting and interpolation using (nonlinear) radial basis functions (RBFs) is proposed. The fit may be constrained very simply to ensure that the resulting surface takes a pre-determined value at one or more specific points. The RBF processor is capable of performing a wide range of complex pattern recognition tasks. The processor, which constitutes a form of nonlinear adaptive filter, behaves like a neural network based on the multilayer, feedforward perceptron model.<
>