Fast layered network data detection based on transmission system properties

Abstract

To drastically accelerate the training process of an M-ary data detector over noisy dispersive channels, based on a radial basis function neural network (RBFNN), data transmission is considered as a whole experiment including the training sequence, the channel, and the adaptive detector. Such a strategy allows only one network basis function center to be updated, leaving the remaining centers to be set in a one-shot fashion prior to data mode. A logarithmic reduction of training time and computation, most beneficial for M>2, is thus possible.