Neural network architectures for signal detection and demodulation

The neural network architecture described is particularly suited to the processing of time-varying signals, signals such as those encountered in discrete binary and M-ary modulation schemes. The basic function of each neuron differs from the majority of conventional neurons in that it acts as a pulse processing unit and is based on an architecture originally postulated by Reiss (1964). The term employed for this type of neuron is pulse discriminant (PD). The author concentrates on the application and performance of simple PD neural networks for the demodulation of binary and M-ary frequency-shift keyed (FSK) modulation schemes