PROGRAMMABLE DIGITAL NEURON USING PULSEWIDTH-CODED DATA

In this paper, a programmable digital neuron architecture using pulsew idth-coded information is presented where the different transfer funct ions, ramp, sigmoid and Gaussian functions, can be generated. The neur on has been extensively simulated. The proposed programmable digital n euron is then applied to the design and analysis of a Gaussian percept ron neural model and its learning algorithm. The modular approach was adopted in the design of the programmable digital neuron to facilitate ease of expansion. (C) 1998 Elsevier Science B.V.