摘要: The backpropagation algorithm is extended to complex domain (CDBP) which can be used train neural networks for the inputs, weights, activation functions, and outputs are complex-valued. Previous derivations of CDBP were necessarily admitting functions that have singularities, highly undesirable. In derivation, derived so it accommodates classes suitable functions. One such function found circuit implementation corresponding neuron given. hardware circuits process sinusoidal signals all at same frequency (phasors). >