A Generic Control Block for Feedforward Neural Network with On-Chip Delta Rule Learning Algorithm

In this paper we propose a method to implement in FPGA a feedforward neural network with on-chip delta rule learning algorithm. For this, we have develop a generic blocks designed in Mathworks´ Simulink environment, capable to generate the signals for controlling the neurons from a neural network. The main characteristics of those blocks is its high reconfigurability that´s makes it suitable for developing of a generic controlling block capable to manage calculus function of neurons from different layers. The properties of the block are set according to the numbers of total layers, number of the neurons from the layers and the number of layer from whom and in different function block parameters windows. The novelty of the proposed method resides in the possibility to design neural networks with on-chip learning only with predefined block systems created in system generator environment. The major benefit of this design methodology result from the possibility to create a higher level design tools used to implement neural networks in logical circuits.