Neural network structure for emulating decision feedback equalisers

Abstract

The emulation of a tuned decision feedback equalizer operating on a noiseless finite impulse response channel by a feedforward multilayer neural network is considered. The similarity between the two structures is exploited to obtain tight bounds on the probability of error for the neural net as a function of the number of layers, using the theory of finite state Markov processes. A class of channels for which exact representation by a neural net of finite complexity is possible is established.